Optimize aligned memory accessors

[RejectKid]

What changed

• generate direct endian-aware reads and writes for byte-aligned, full-width Memory<byte> fields
• support signed and unsigned integral fields, enums, both bit orders, nonzero byte offsets, and read-only getters
• add focused UInt32 and UInt64 generated-accessor benchmarks
• include the new benchmarks in the original-performance regression harness
• add randomized generator correctness coverage

Why

Generated accessors for aligned memory fields previously routed through the general-purpose bit primitives. Because field width, offset, and bit order are compile-time constants, the generator can emit direct BinaryPrimitives operations and avoid the general extraction/write path.

Performance

Paired BenchmarkDotNet MediumRun results on AMD Ryzen 9 9900X / .NET 10:

Operation	Before	After	Improvement
UInt32 get	0.8491 ns	0.7776 ns	8.4%
UInt64 get	0.8433 ns	0.7925 ns	6.0%
UInt32 set	0.9716 ns	0.8385 ns	13.7%
UInt64 set	1.0299 ns	0.8697 ns	15.6%

All measured operations remain allocation-free.

Validation

• dotnet test BitsKit.Tests/BitsKit.Tests.csproj -c Release --no-restore
• 4,588 tests passed on .NET 8
• 4,588 tests passed on .NET 10
• dotnet build BitsKit.sln -c Release --no-restore
• git diff --check