perf: add HybridUint::read fast path for msb_in_token == 0

[hjanuschka]

Carved out from the ongoing perf/autoresearch stream after PR #705 split work.

Change:

• add a specialized fast path in HybridUint::read for the common msb_in_token == 0 case.

Why:

• avoids general high-bit reconstruction on a hot path
• keeps existing behavior

[github-actions]

Benchmark @ 3ba6a16

MULTI-FILE BENCHMARK RESULTS (8 files)
  CPU architecture: x86_64
  WARNING: System appears noisy: high system load (2.63). Results may be unreliable.
Statistics:
  Confidence:               99.0%
  Max relative error:        3.0%

Comparing: 7cf3a662 (Base) vs 38065842 (PR)

File	Base (MP/s)	PR (MP/s)	Δ%
bicycles.jxl	7.630	7.639	+0.12% ±2.9%
bike.jxl	25.526	25.485	-0.16% ±1.2%
delta_palette.jxl	6.228	6.199	-0.46% ±1.2%
green_queen_modular_e3.jxl	8.422	8.386	-0.43% ±0.3%
green_queen_vardct_e3.jxl	26.003	26.088	+0.33% ±1.2%
lz77_flower.jxl	3.444	3.442	-0.04% ±1.2%
patches_lossless.jxl	3.276	3.273	-0.06% ±0.5%
sunset_logo.jxl	2.893	2.852	-1.43% ±1.3%

[veluca93]

I'd say that since perf is neutral the additional code complexity is not worth it, so closing it for now.