Add opt-in compact storage for Animate timelines

[Tutez64]

Summary

This PR adds an opt-in swf_compact_animate_timelines define that stores
Animate timeline events in compact numeric buffers instead of eagerly creating
an object graph containing one AnimateFrame, AnimateFrameObject, and
optional matrix, color-transform, filter, and array objects per serialized
event.

The compact representation:

• stores frame boundaries as offsets into a flat Vector<Float>;
• keeps non-numeric values in a separate sparse reference array;
• decodes event properties directly during playback;
• updates native transforms without allocating temporary Matrix objects;
• preserves the existing Animate asset format and standard playback behavior.

The legacy representation remains the default.

This is opt-in because code that accesses AnimateSpriteSymbol.frames directly
will not receive the eagerly expanded legacy metadata.

Motivation

The expanded timeline metadata remains reachable for the lifetime of an
Animate library. Large libraries can therefore retain hundreds of thousands of
small managed objects, increasing both heap usage and the amount of graph
traversal performed by full garbage collections.

Deterministic benchmark

SWFAnimateTimelineMemoryRepro.zip
It parses and retains 400 sprites containing 300 frames and
eight events per frame, for a total of 960,000 timeline events.

Three alternating runs on Linux x86_64 with Haxe 4.3.6 produced:

Metric	Legacy	Compact
Parse time	492–543 ms	497–500 ms
Retained hxcpp heap	218.6 MiB	93.5 MiB
First full GC	34–38 ms	4 ms
Median forced compacting GC	66-79 ms	1 ms

The retained heap decreased by 57%. The benchmark intentionally amplifies
timeline metadata costs; application results depend on the amount and
complexity of loaded animation data.

Real-world validation

These values should be considered indicative rather than a controlled benchmark:

Metric	Legacy	Compact
Marking time per GC	39.57 ms	27.13 ms
Full-scan marking average	109.66 ms	24.82 ms
Full-scan marking maximum	233.78 ms	40.20 ms
Maximum GC pause	322.16 ms	140.15 ms
GC time fraction	0.792%	0.636%

At a comparable heap census, the compact run contained approximately 4,171
AnimateTimelineData objects instead of 40,949 AnimateFrame and 63,076
AnimateFrameObject objects. Retained Matrix instances decreased from
138,766 to 78,802.

Tests

• 57/57 Neko tests pass with the legacy representation.
• 57/57 Neko tests pass with swf_compact_animate_timelines.
• Native C++ builds and extended runtime sessions completed successfully.
• Timeline playback tests cover creation, updates, destruction, transforms,
  color transforms, filters, names, visibility, and frame labels.

Integration notes for pending PRs

This branch is intentionally based directly on upstream/master and does not include other pending changes. Depending on merge order, the following adaptations must be preserved when resolving conflicts:

• Timeline instance cleanup: compact timeline instances must set __removedChildCleanupDelay to Math.POSITIVE_INFINITY when created.
• Timeline masks: masks from compact timelines must also be marked through __markTimelineMask. This can be handled in the shared instance-arrangement path.
• Clip-depth deduplication: compact instance reuse must include clipDepth in its compatibility check.
• Cached instance fields: preserve the reset-dependent filter clearing behavior when merging the shared display-object update code.
• Hardware bitmap cache (Add opt-in hardware cache for Animate bitmaps #56): when compact timelines are enabled, the readability scan must inspect compactTimeline.objects, since the legacy frames array is intentionally left empty.

These integrations are already implemented and tested together on the downstream master branch; they are omitted here only to keep this PR independent and reviewable.

Note: until 159db83 from #55 is merged, the Haxe 3 CI jobs may still fail earlier in AnimateTimeline.hx, independently of this PR.