fix(runtime): make hard-sync and preload boundaries inclusive at clip end

[calcarazgre646]

Follow-up to #1173, which made the audio clock and syncRuntimeMedia treat t = end as active to match the visibility change from #1166. Two sites in the runtime still use the exclusive end:

hardSyncAllMedia (init.ts) - observable bug

hardSyncAllMedia skips elements when timeSeconds >= end, so pausing exactly at a clip's end boundary skips the seek for that clip. This matters because the skip is not covered by any other path: syncRuntimeMedia deliberately avoids strict/force sync while a video is playing (decoder-reset cost, see the drift-correction tiers in media.ts), so drift under 0.5s on a playing video is only ever corrected by the play/pause hard sync. Pause exactly at t = end and the paused video shows its drifted frame instead of the final one, while the element is still visible (inclusive since #1166).

getClipsInWindow (mediaPreloader.ts) - consistency only

The preloader's active predicate also used the exclusive end. To be transparent: this one has no observable behavior change, because a clip at exactly t = end stays in the preload window via the lookbehind term on the next line. Updated so the predicate matches the runtime-wide definition of active rather than leaving the last exclusive comparison in the runtime.

Tests

Two tests in init.test.ts driving the real player through the manual-RAF harness:

• pause exactly at t = end hard-syncs the element to its final frame (fails without the fix: currentTime stays at the drifted value)
• pause past t = end does not seek the element (guards the upper bound)

The fixture pins drift at 0.3s, below the 0.5s hard-sync tier and above the force-sync threshold, so the assertion isolates hardSyncAllMedia from syncRuntimeMedia's own corrections.

Full core suite green (73 files, 1439 passed).

[miguel-heygen]

Clean fix — two one-character changes (`>=` → `>`, `<` → `<=`) aligning the last two exclusive-end sites with the inclusive-end contract from #1173/#1166.

init.ts `hardSyncAllMedia` — the observable bug. Changing `timeSeconds >= end` to `timeSeconds > end` means pausing exactly at a clip's end boundary now seeks the video to its final frame instead of skipping it. The test proves the fix: clip [2,5], pause at t=5, video.currentTime lands on 3 (correct final relative time) instead of staying at the drifted 3.2.

mediaPreloader.ts `getClipsInWindow` — consistency. `timeSeconds < clip.end` → `timeSeconds <= clip.end` matches the inclusive semantics everywhere else. No user-visible impact but prevents future confusion.

Tests are solid: boundary fixture with deliberate sub-threshold drift (3.2 vs 3.0 — under the 0.5s hard-sync tier), one test at exact boundary (t=end, expects correction), one test past boundary (t=end+0.5, expects no correction).

Ship it.