Optimize char deserialization with manual UTF-8 decoder

[tanmay4l]

addresses the TODO comment at line 247-250 which noted: "Could implement a manual decoder that avoids UTF-8
validate + chars() and instead performs the UTF-8 validity checks and produces a char directly. Some quick
micro-benchmarking revealed a roughly 2x speedup is possible."

Changes

Before:

let str = core::str::from_utf8(buf).map_err(invalid_utf8_encoding)?;
let c = str.chars().next().unwrap();

After:
- Manual UTF-8 decoding for 2-4 byte characters using bit masks
- Inline validation of continuation bytes (must be 10xxxxxx)
- Overlong encoding validation (3-byte: >= U+0800, 4-byte: >= U+10000)
- Surrogate validation (rejects U+D800..U+DFFF)
- Out of range validation (rejects > U+10FFFF)

[kskalski]

You mention using a microbenchmark? Does it make sense to include it in the PR (e.g. add to wincode/benches) and get the comparison numbers in the PR description?

[kskalski]

Thanks, I used the code and did a few other changes on top of it including the benchmark too in #187