feat: batch video upscaling with folder upload, directory save, and result comparison (#1051)

* feat: support batch video upscaling with folder upload and directory save

- Add multi-file selection and entire folder upload (webkitdirectory)
- Support drag & drop of both files and folders via FileSystemEntry API
- Add save directory picker using File System Access API (showDirectoryPicker)
- Process videos sequentially with per-file progress and status tracking
- Add right-side result list with click-to-compare functionality
- Add per-video duration and total batch duration display
- Add retry button for failed items with error detail display
- Remove auto-play on comparison viewer load
- Fix worker error handling: handle "error" command to reject stale promises
- Fix event listener accumulation in video comparison (AbortController)
- Fix XSS risk by escaping HTML in all innerHTML injections
- Fix URL revocation timing in download to prevent large file truncation
- Extract writeToSaveDirectory and truncateName helpers to reduce duplication
- Add Usage Guide panel explaining single/multiple/folder upload options

Files changed:
  docs/upscaler/index.html  (+1110/-181)
  docs/upscaler/upscaler.js (+110)

* fix: prevent auto-play and resource consumption during upscaling

- Set preload="metadata" on preview and comparison video elements
- Explicitly pause preview video after setting src in showVideoPreview
- Remove preview video seek sync from batch processing progress callbacks
  (was causing continuous decode during upscaling, wasting GPU resources)
- Remove preview video seek sync from retry progress callback
- All videos now require manual play; no background resource usage

* fix: improve upscaled video quality with higher bitrate and codec support

- Raise base bitrate from 2.5 Mbps (25e5) to 8 Mbps (8e6) for standard and
  firefox processors, preventing quality loss when source videos have high
  bitrates (e.g. 720p @ 16.7 Mbps was producing 1440p @ 10 Mbps, smaller
  than the original file)
- Double the bitrate for mediabunny processor (X.n1 * 2) to account for the
  4x pixel increase from 2x upscaling
- Add H.264 Level 5.1 (avc1.640033) and Level 5.2 (avc1.640034) codecs to
  the fallback lists, since Level 4.0 (avc1.640028, ~24.6 Mbps) cannot
  support the higher bitrates for 1440p+ outputs
- Add latencyMode: "quality" to VideoEncoder configs for both standard and
  firefox processors, prioritizing encoding quality over latency (suitable
  for offline processing)

* feat: add video data comparison panel for original vs upscaled metrics

- Add a comparison data table below the file info bar showing 6 metrics:
  Resolution, File Size, Duration, Est. Bitrate, Pixels, Process Time
- Display per-metric change percentage with color-coded indicators:
  green (↑) for improvements, red (↓) for regressions, gray for neutral
- Estimated bitrate calculated from file size × 8 / video duration
- Table renders after both original and upscaled video metadata loads
- Auto-hide panel on reset

* fix: make video data comparison panel full-width and responsive

- Move comparison data table from inside the left column to the full-width
  area below the left/right layout, so it is no longer constrained by the
  comparison viewer column width
- Remove table-fixed colgroup and overflow-x-auto that caused horizontal
  scrolling on smaller screens
- Table now uses w-full auto-layout, displaying all 4 columns (Metric,
  Original, Upscaled, Change) without requiring horizontal scrolling

---------

Co-authored-by: rionzhao <rionzhao@tencent.com>

assets/sample/danmuku-v3.xml

@@ -0,0 +1 @@
+<?xml version="1.0" encoding="UTF-8"?><i><chatserver>chat.bilibili.com</chatserver><chatid>580770634</chatid><mission>0</mission><maxlimit>1500</maxlimit><state>0</state><real_name>0</real_name><source>k-v</source><d p="481.89400,1,25,16777215,1653692078,0,70e36298,1062005140992593408,11">可以去太湖捞捞试试（狗头保命</d><d p="265.37800,1,25,16777215,1653616680,0,88afabc0,1061372654017566976,11">好了，今天就到这，我感觉生物钟启动了，我要睡了</d><d p="777.55400,1,25,16777215,1653575476,0,9bcc00a2,1061027009955725568,11">当然是sha256(sha256(pw)+salt)</d><d p="13.88600,1,25,16777215,1653570775,0,966e6348,1060987581535333376,11">Swift～～～</d><d p="454.63400,1,25,16777215,1652717912,0,6d2da793,1053833248746068480,11">搞个共享数据库，像挖矿一样每个人算好的都可以上传</d><d p="177.13900,1,25,16777215,1652453333,0,fb4f660b,1051613793517396736,11">有点懵，但感到很震撼</d><d p="22.57900,1,25,16777215,1652430212,0,bdb52a9e,1051419843691098880,11">哈希值</d><d p="58.06700,1,25,16777215,1652368788,0,e91e8413,1050904578150762496,11">有了粉碎机，家里就只剩下粉碎机了</d><d p="275.65900,5,25,16777215,1652354865,0,397b3aab,1050787785968982016,11">《我知道了》</d><d p="774.77900,1,25,16777215,1652256952,0,542886c4,1049966433997495040,11">奥卡姆剃刀</d><d p="352.26200,1,25,16777215,1652247365,0,139fd1a2,1049886011934382080,11">加盐只是通过不确定的盐来隐藏确定的原值，增加彩虹表的制作难度</d><d p="395.65900,1,25,16777215,1652169656,0,e2308d98,1049234139607279872,11">哈希冲突</d><d p="346.72900,1,25,16777215,1652112430,0,54ba448a,1048754095986328064,11">《彩虹表》</d><d p="653.60600,1,25,16777215,1652101305,0,5ec1b46c,1048660766959520768,11">哈哈，这个可以使git库瞬间挂掉</d><d p="461.75200,1,25,16777215,1652101091,0,5ec1b46c,1048658979179723264,11">做个超级字典，把最常用的组合记录下来就行了</d><d p="94.11200,1,25,16777215,1652100683,0,5ec1b46c,1048655552173002240,11">哈希根本不是加密，只是摘要算法</d><d p="386.19500,1,25,16777215,1652097650,0,c4cc4296,1048630114516051968,11">组合数学开始了</d><d p="414.69100,1,25,16777215,1652084321,0,fafcfbb0,1048518297794048256,11">设计者自己找不到（doge</d><d p="59.62300,5,25,16707842,1652052664,0,971041b8,1048252744781197824,11">我会!</d><d p="445.99300,1,25,16777215,1651998604,0,53d75164,1047799249896694016,11">哎嘿 很勇哦</d><d p="641.24300,1,25,16777215,1651991247,0,ed4717ed,1047737536325133312,11">但是加盐还是有用</d><d p="408.97200,1,25,16777215,1651977907,0,4bd9a2d7,1047625636933279232,11">可以存在一个笼子里一只鸽子也没有</d><d p="511.84900,1,25,16777215,1651972503,0,f0826639,1047580298436422144,11">哈希冲突</d><d p="520.96700,5,25,16777215,1651967260,0,27acb016,1047536317854342656,11">列巴的列（补充</d><d p="558.75200,1,25,16777215,1651940306,0,860de6bd,1047310216540149504,11">《模仿游戏》来了</d><d p="505.15000,1,25,16777215,1651940233,0,860de6bd,1047309599239544064,11">电影《模仿游戏》</d><d p="488.98600,1,25,16777215,1651927542,0,44f3ad45,1047203142191739648,11">俺姆昂鸽瑞！</d><d p="586.48600,1,25,16777215,1651849723,0,29026d4b,1046550345196703488,11">烫烫烫哈哈哈哈哈哈！</d><d p="102.05200,1,25,16777215,1651839216,0,1336ac64,1046462209011973120,11">王小云，指数级缩短了MD5的碰撞时间</d><d p="55.82000,1,25,16777215,1651825875,0,38443e73,1046350293337386752,11">一半我做的登录都会生成一个密码盐，存在本地</d><d p="681.33200,1,25,16777215,1651819803,0,fc3f2a25,1046299363606827264,11">加盐也对付不了碰撞 只能对付彩虹表</d><d p="18.26500,1,25,16777215,1651816441,0,cf2bea01,1046271154899279872,11">A(欸)P(劈)P(劈)</d><d p="567.29500,1,25,16777215,1651812016,0,5309839c,1046234041289880832,11">你错了，图灵对这个并不熟</d><d p="504.07300,1,25,16777215,1651806653,0,58c10bda,1046189050064611584,11">加盐</d><d p="350.22000,1,25,16777215,1651798844,0,d3cbe582,1046123548323639040,11">所以要加盐</d><d p="59.55900,1,25,16777215,1651793262,0,dc9b4d74,1046076722316952576,11">哈希有很小很小的概率把两个不一样的东西映射成一个</d><d p="468.25000,1,25,16777215,1651762009,0,1e90bb3e,1045814547245588736,11">量子计算：这个我熟</d><d p="565.29300,1,25,16777215,1651722446,0,1b70a4,1045482667682264064,11">图灵说:这个我熟</d><d p="652.84300,1,25,16777215,1651713497,0,9e19237c,1045407604815323904,11">图片隐写</d><d p="391.57200,1,25,16777215,1651622858,0,d78de8a,1044647266629422336,11">吾生而有崖</d><d p="273.57700,1,25,16777215,1651509100,0,c8ed82e9,1043692993388749056,11">《也可以100行以内》</d><d p="494.74000,1,25,16777215,1651405107,0,8262fe93,1042820639116808192,11">数据库！</d><d p="753.23500,1,25,16777215,1651385801,0,ea9eeeda,1042658687317412864,11">火速换RSA!</d><d p="747.76900,1,25,16777215Line truncated