Bugfix/jpeg overflow guard

[IfkumRfnl]

Summary

Fixes #324 by returning Unsupported before the decoder reaches overflowing fixed-point arithmetic.

Details

• Checks JPEG dequantization multiplication and returns Image.ReadError.Unsupported on overflow.
• Rejects unsupported 16 bit quantization tables in the current 8-bit JPEG decode path.
• Adds a conservative input bound before each fixed point IDCT pass so existing i32 arithmetic cannot overflow.
• Propagates IDCT errors through the JPEG and pipeline.
• Tightens crafted overflow tests to expect Image.ReadError.Unsupported.

Small note: I didn't go with widening the arithmetic types since the SIMD operations underneath are currently built around i32, so I think for now just not supporting these kind of images is better.

[IfkumRfnl]

@mlarouche could you review this when you have a chance? Companion fixture PR: zigimg/test-suite#39

[stone-d-chen]

thanks for the fix!