docs: bulk authoring + track-selection guidance (#43 Phase 2)

[jdwit]

Phase-2 of #43. Skill polish only - no CLI change, no new command or flag. Builds on the Phase-1 primitives (videos captions/transcript, profile brand) shipped in #52.

What changed (skills/ytstudio/SKILL.md)

Bulk authoring subsection - the recommended agent loop to backfill metadata across many videos:

1. Select candidates cheaply with one videos list -o json pass, filtering weak-metadata fields client-side (description == "", tags == [], null language fields, empty localizations).
2. Gate the pricey videos transcript behind a cheap videos captions check - skip the transcript entirely for videos with no usable track.
3. Pull grounding + current state only for the survivors.
4. Draft on-brand, preview each videos update as a dry-run.
5. Batch-confirm the whole set with the user, then --execute each.

Quota-aware batching - real arithmetic against the 10,000 units/day budget: ~350 units per fully-authored video (the transcript command itself lists then downloads, so the gate's list call is paid twice when a track exists), so ~28 videos exhaust a fresh budget; estimate candidates x 350 and warn before a run that could hit quotaExceeded.

Expanded language/track selection - prefer standard over ASR, pass --lang explicitly on multi-language channels (without it transcript auto-picks across all tracks), and fall back to videos captions / tell the user when a track is missing or not downloadable.

Phase 3 (external ASR/STT) and any new CLI primitive stay out of scope.

Scope & verification

Designed and reviewed with the same multi-agent approach as Phase 1: a design workflow confirmed this is expressible with existing primitives (no new CLI warranted), and an adversarial review over the diff surfaced two issues, both fixed here:

• A --has-localization example was dropped - the flag keeps videos that already have a localization, the inverse of the gap-backfill candidates the section is about.
• The per-video quota estimate was corrected from ~300 to ~350 units (transcript internally re-lists captions), so the "~28 videos" warning errs on the safe side.

No CLI code changed, so the generated references/reference.md is untouched and tests/test_skill_reference.py stays green. Full suite: 338 passed; SKILL.md is ASCII-only and every command/flag/field referenced is verified against the live CLI.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.03%. Comparing base (360a106) to head (bf16661).

Additional details and impacted files

@@           Coverage Diff           @@
##             main      #53   +/-   ##
=======================================
  Coverage   85.03%   85.03%           
=======================================
  Files          15       15           
  Lines        2660     2660           
  Branches      393      393           
=======================================
  Hits         2262     2262           
  Misses        264      264           
  Partials      134      134           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.