chore(scripts): render cli_interactive tapes in parallel

[bearomorphism]

Description

Follow-up to PR #2009 (deterministic gif regeneration) that makes local iteration faster:

1. Parallelism in scripts/gen_cli_interactive_gifs.py. Tapes now render through a ThreadPoolExecutor with the default capped at min(len(tapes), 4); -j N overrides. Errors are reported per tape with a final summary instead of aborting on the first failure. The poe task doc:screenshots still calls gen_cli_interactive_gifs() with no args (zero-arg compatible signature), so nothing changes for the standard render everything path.

2. Per-tape workdir in shared/base.tape + shared/cleanup.tape. Both files used to hardcode /tmp/commitizen-example, which would race once multiple tapes ran concurrently. Replaced with mktemp -d "${TMPDIR:-/tmp}/commitizen-example.XXXXXX" saved in a per-tape $WORKDIR shell variable and cleaned up at the end. Rendered gifs are unchanged (the setup runs inside Hide).

Checklist

• I have read the contributing guidelines

Was generative AI tooling used to co-author this PR?

• Yes (please specify the tool below)

Generated-by: GitHub Copilot CLI (claude-opus-4.7-1m-internal) following the guidelines

Code Changes

• Run uv run poe all locally to ensure this change passes linter check and tests
  • poe lint clean (ruff + mypy). Same pre-existing Windows-only flakes as PR ci(docspublish): make cli_interactive gif regeneration deterministic #2009 (gpg setup + a parallel-test race), unrelated to this change.
• Manually test the changes:
  • python scripts/gen_cli_interactive_gifs.py --help -- argparse output renders.
  • vhs validate passes on all 5 tapes after the shared/ patch.
  • Full vhs render not run locally (native Windows vhs hangs on ttyd+WSL bash; same caveat as PR ci(docspublish): make cli_interactive gif regeneration deterministic #2009).
• Update the documentation for the changes
  • N/A -- behaviour of uv run poe doc:screenshots is unchanged.

Expected Behavior

• uv run poe doc:screenshots continues to render all five interactive gifs (now in parallel, with the same outputs).
• python scripts/gen_cli_interactive_gifs.py [-j N] renders all tapes with up to N parallel workers (default min(len(tapes), 4)).
• Parallel runs no longer race on a shared workdir, because each tape now operates in its own mktemp -d directory.

Steps to Test This Pull Request

# Lint
uv run poe lint

# Argparse
python scripts/gen_cli_interactive_gifs.py --help

# Tape syntax still valid
cd docs/images && vhs validate bump.tape commit.tape init.tape shortcut_default.tape shortcut_custom.tape

# (optional, Linux/WSL) End-to-end render
uv run python scripts/gen_cli_interactive_gifs.py        # all 5 in parallel
uv run python scripts/gen_cli_interactive_gifs.py -j 1   # serial
uv run poe doc:screenshots                               # poe task still works

Additional Context

Depends conceptually on but does not require PR #2009 (deterministic gifs). The two PRs touch disjoint files except for shared/base.tape and shared/cleanup.tape, which only PR B modifies.

[github-actions]

🔍 Commitizen bump preview

Merging this PR will produce the following bump:

bump: version 4.16.3 → 4.17.0
tag to create: v4.17.0
increment detected: MINOR

[github-actions]

🔍 Commitizen bump preview

No commits in this PR are eligible for a version bump.

[bearomorphism]

Slimmer rewrite at 68fadd8: PR diff dropped from +151/-29 to +100/-29 by inlining helpers (_render_one, _resolve_tapes, _default_workers, _build_parser, main) and replacing the CPU-aware worker default with min(len(tapes), 4). Public gen_cli_interactive_gifs(tape_names=None, max_workers=None) signature, CLI flags, preflight VHS check, and per-tape error reporting are unchanged. Script went from 37 to 94 lines (was 159).

[github-actions]

🔍 Commitizen bump preview

No commits in this PR are eligible for a version bump.

[github-actions]

🔍 Commitizen bump preview

No commits in this PR are eligible for a version bump.

[bearomorphism]

Trimmed further at daffcb6: dropped the tape-name filtering (CLI positional + by_stem resolution + dedup loop) per discussion. The script now just runs every *.tape in parallel; -j is the only remaining flag. PR diff is now +71/-29 (was +151/-29), script is 68 lines (was 159, original 37). Re-titled accordingly.