render: Flush GPU work periodically during cache entry processing to fix OOM error

[Bubby4j]

(This is my first Ruffle contribution, please review critically - this was AI generated but did fix the issue I'm seeing)

When rendering objects with extremely large dimensions whose cacheAsBitmap is skipped (in my case 6.7M × 6.7M pixel objects), their children's bitmap caches are all submitted in a single GPU frame without any intermediate flushes. This accumulates GPU resources until the device runs out of memory.

Root Cause
In submit_frame, the cache_entries loop renders all cached bitmaps sequentially without calling maybe_flush(). The existing maybe_flush() mechanism (which periodically submits GPU work to prevent OOM) was only being invoked during command rendering, not during cache entry rendering.

Fix
Call self.active_frame.maybe_flush() inside the cache_entries loop, matching the pattern already used elsewhere to bound GPU resource accumulation within a single frame.

Testing
Tested with a SWF file that previously triggered a WGPU OOM crash due to a ~6.7 million pixel object. The crash no longer occurs.

[kjarosh]

LGTM, that sounds reasonable. Thank you!

[kjarosh]

I'll mark it as newsworthy in advance, but we'll have to see if we get fewer OOM reports after merging it.

[kjarosh]

Just a note for the future: if this causes performance regressions, I think we should tune MAX_DRAWS_PER_FLUSH.