Add MPMask.getAsImageBitmap() for Web segmentation

[a2105z]

Summary

MPImage already has getAsImageBitmap(). MPMask (confidence / category masks from ImageSegmenter, Interactive Segmenter, PoseLandmarker) did not. Segmentation effects on Insertable Streams / BreakoutBox / Web Workers want:

new VideoFrame(result.confidenceMasks[0].getAsImageBitmap(), { timestamp: frame.timestamp });

Workarounds today are either getAsFloat32Array() (GPU→CPU readPixels) or poking mask.canvas['GLctxObject']. Schmidt noted this was already on the internal list in 2023.

Fixes #4798

API

mask.hasImageBitmap(): boolean
mask.getAsImageBitmap(): ImageBitmap  // synchronous

• GPU path: sample the RED channel, write grayscale RGBA vec4(m, m, m, m) to the OffscreenCanvas, transferToImageBitmap() — same sync GPU path as MPImage, no float readback.
• CPU path: upload once to a texture, then the same transfer.
• Requires the task to be created with an OffscreenCanvas (same limitation / error as MPImage.getAsImageBitmap()).
• Cached on the MPMask; close() frees an owned bitmap.

Test plan

• mask.test.ts: hasImageBitmap() after conversion; grayscale pixels from Uint8Array and WebGLTexture; cache identity; clone owns a distinct bitmap; throws without a canvas
• bazel test //mediapipe/tasks/web/vision/core:mask_test (browser; skipped in Node like the existing mask tests)

@schmidt-sebastian @kuaashish @HarishPermal @tyrmullen @kalyan2789g @akashvverma1995 @ayushgdev @lu-wang-g