fix(core): capture pages load only the clips on screen - #7402

Request changes at 0376faa0. The code looks right to me. Two gate items are blocking.

Blocking

1. The body has no fails without the fix: line. This is a fix PR, so it needs one line naming the test that fails without the fix. The Test plan does describe the failure, but not on the required line. Add, for example:
   fails without the fix: packages/core/src/runtime/entry.test.ts › "in a hidden capture page only the clip on screen loads, and the one a seek left is freed"
   I checked the claim locally. With only init.ts reverted to the merge-base 7ad2be96, that test fails with expected [ true, true, false ] to deeply equal [ true, false, false ]. At the head it passes, and so does the rest of entry.test.ts (57/57).
2. The Comments workflow fails on this head (comment-ratchet, "A package file's comment share may not rise"):
   • packages/core/src/runtime/init.ts: 900 → 901 comment lines. The new // A capture page seeks and never plays… line is the one that tips it.
   • packages/core/src/studioPreviewMark.ts: 0 → 2 comment lines (the JSDoc on capturePageIn).
     Fix: cut them, or move the explanation into the name. For example, capturePageIn already says most of it, and the rule could sit in a named const such as const loadsAhead = !capturePage.

Checked, no issue

• capturePageIn matches the marker read it replaces in timedClipHide.ts (same meta[name=…] selector and content === "capture"). The pages that set the marker are the studio-server preview route and checkBrowser.
• The live preview and the player don't change: every new branch is behind capturePage.
• On a capture page, releasing with force = true only frees a clip that was already decided and has now left the screen, which is what the PR intends.

Non-blocking

• As the body notes, hyperframes check hasn't been measured end to end. Now that nothing is prefetched, every check seek onto a new cut starts a cold load. That's worth one timing run on a heavily cut film, to confirm check's wall time doesn't regress where memory improves.

I'll re-review once both blockers are fixed on a new head.

Approve at c6ab3a66. Both blockers from my earlier review are fixed.

• fails without the fix: line: it's in the body now and names entry.test.ts › "in a hidden capture page only the clip on screen loads, and the one a seek left is freed". I checked it again at this head. With only init.ts at the merge-base 7ad2be96, the test fails with [true, true, false] vs [true, false, false]. At the head, entry.test.ts passes 57/57.
• Comment ratchet: the comments are gone, and the rule now lives in const loadsAhead = !capturePageIn(document). Between 0376faa0 and this head the only change is that rename, with logic identical under De Morgan (capturePage → !loadsAhead). The Comments workflow now passes.
• CI on this head: all workflows passed (CI, Comments, regression, preview-regression, Windows render, Player perf/Safari perf/Safari audio, CodeQL, PR captures). The cancelled duplicates were superseded by later runs on the same head.

Still non-blocking, as before: one end-to-end hyperframes check timing run on a heavily cut film, to confirm wall time doesn't regress now that nothing is prefetched.