Add generated parser benchmarks

[sebastienros]

The README benchmark data was stale and did not consistently cover Parlot's source-generated parsers. This adds generated parser variants and refreshes the published measurements on .NET 10.

Changes

• Add source-generated JSON benchmarks for big, deep, long, and wide documents.
• Include generated expression and JSON cases in the aggregate benchmark suite.
• Cache and benchmark the generated email parser alongside Fluent and compiled variants.
• Refresh README expression, JSON, and email results and update the documented dependency versions.
• Correct the compilation guidance now that recent Fluent hot-path optimizations have narrowed the compiled-vs-Fluent gap.
• Resolve implementation assemblies from reference-assembly paths and preserve the net10.0 Parlot assembly in BenchmarkDotNet's flattened output directory.

The assembly handling is necessary because BenchmarkDotNet redirects project outputs into one directory; without it, the generator can load reference assemblies during its nested build or the netstandard2.0 Parlot assembly can overwrite the net10.0 runtime assembly.

Validation

• All 46 README benchmark cases completed on net10.0.
• Source generator tests passed (248 tests).
• Release benchmark project build succeeded.