[rudy-07]

What

Updates the warning emitted by hyperframes audio --read when stem separation is skipped at local or fast quality. The warning now describes what content is omitted instead of instructing the caller to pass --quality best.

Why

Fixes the remaining item of #7204.

When audio --read runs under --quality local or --quality fast, the CLI previously emitted:
not separated into stems: quality <quality> keeps audio on this machine; --quality best separates it through HeyGen

When the CLI is invoked non-interactively or wrapped by host tools and agents that cannot supply flags, directing the caller to pass --quality best provides an action they cannot execute. Stating what is missing ("music, and effects under speech, are not read") communicates the read status directly without prescribing flags.

Related work

Fixes #7204.
Follows #7345 (which resolved item 1 of #7204 by adding --layer).

How

In packages/cli/src/audio/read/read.ts, update stemsFor() to push not separated into stems: music, and effects under speech, are not read when quality !== "best". Updated the corresponding regression assertion in packages/cli/src/audio/read/read.test.ts.

Test plan

fails without the fix: packages/cli/src/audio/read/read.test.ts > keeps audio on the machine by default, and reads neither voice nor effects from a bare mix expects the new warning text.

• All 17 read.test.ts tests pass.
• Linter and formatter checks clean: bun run lint and bun run format:check.
• Fallow and typecheck checks clean across the workspace.
• Comment citation and ratchet checks clean: node scripts/check-comment-citations.mjs and node scripts/comment-ratchet.mjs.