Preserve exact fuzzy percentage for fully matched code

[itsgrimetime]

Fully matched reports can show a fuzzy percentage such as 99.999954 even when matched_code == total_code and matched_code_percent == 100. The weighted f32 accumulator can drift above or below the exact sum while accumulating function and unit scores.

Return exactly 100% when the integer matched and total code byte counts are equal. The shared normalization method covers unit, category, and overall measures, preserves the existing empty-code behavior, and leaves partially matched weighted scores unchanged. No report schema changes are needed.

Adds regression coverage for accumulated sums on both sides of 100%, aggregation of 1,130 fully matched units, a near-complete partial match, and empty code.

Validation: cargo test passes (54 tests); git diff --check passes. The required nightly formatting and Clippy commands could not run because nightly/rustup is unavailable locally. A stable Clippy attempt stops on the existing clippy::useless_borrows_in_formatting lint being unknown to the installed toolchain; stable formatting also cannot honor the repository's nightly-only options.