Improve codec batching and Metal kernels

[jcwal1516]

Summary

• Improve JPEG batch decoding, grouped Metal kernels, and pooled output reuse.
• Update HT cleanup and IDWT Metal paths, with parity and rounding coverage.
• Add benchmark evidence and update the SIMD dependency.

Validation

• git diff --check passed.
• Local tests were not run.
• Metal quick validation will be dispatched per the repository's PR guidance.