Bound surface reconfiguration retries

[jackwalkerlabs]

Closes #460.

A persistent Suboptimal, Outdated, or Lost result currently reconfigures the surface and re-enters an unconditional loop. If reconfiguration cannot correct the status—such as while the cached size is stale—render_with never returns and the event loop cannot deliver the resize that would recover it.

This restores the bounded behavior from before the wgpu 29 update:

• reconfigure and retry acquisition once;
• return Error::Surface if a recoverable status persists;
• retain the existing skip behavior for Occluded/Timeout and validation error handling;
• regression-test that only one reconfiguration is performed before returning the error.

Validation:

• cargo fmt --all -- --check
• cargo check --workspace
• cargo doc --workspace --no-deps
• cargo clippy --workspace --tests -- -D warnings
• cargo test --workspace
• CI-equivalent WASM and Android cargo check/Clippy commands for all three cross targets
• git diff --check

The Linux/X11 driver hang is not available on this macOS host; the source-level failure is deterministic, and the regression exercises the bounded recovery policy without requiring a real GPU surface.

[parasyte]

Thanks. I opened #462 with my preferred fix to the unbounded loop problem.

[jackwalkerlabs]

Thanks for taking this over. #462 covers the bounded behavior with the implementation you prefer, so I am closing this one.