perf(csharp): add csharp benchmarks and optimize hot serialization paths

[chaokunyang]

Why?

• Add a first-class C# benchmark suite aligned with existing benchmark datasets and serializers.
• Improve C# runtime hot paths while keeping benchmark comparisons fair.
• Fix enum behavior so undefined numeric enum values can round-trip instead of failing.

What does this PR do?

• Adds benchmarks/csharp benchmark harness (Program.cs, models, serializer adapters, run.sh, and report generator) for fory, protobuf-net, and MessagePack.
• Benchmarks cover Struct, Sample, MediaContent and list variants for both serialize/deserialize.
• Uses directly serializable benchmark models for all three serializers (removed protobuf conversion adapter overhead from timed path).
• Updates benchmark report generation to keep throughput and size reporting separate, and adds a C++-style Serialized Data Sizes (bytes) matrix.
• Optimizes C# runtime hot paths: reusable ByteWriter/ByteReader, span-based string encode/decode path, and cached allowed wire-type sets.
• Improves generated/runtime schema handling: CheckStructVersion is wired through contexts/generated serializers with cached no-trackRef schema hash.
• Optimizes collection and enum hot paths: sealed collection element types avoid redundant type-info writes; enum serializer uses cached maps and preserves unknown numeric enum values on read.
• Simplifies config surface by removing skip-root/reflection-fallback flags and setting default MaxDepth to 20.
• Adds/updates tests for schema version behavior and undefined enum round-trip.

Related issues

#1017 #3387
#3397

Does this PR introduce any user-facing change?

• Does this PR introduce any public API change?
• Does this PR introduce any binary protocol compatibility change?

Benchmark

• Command:
  • cd benchmarks/csharp && ./run.sh --duration 3 --warmup 1
• Selected results (ops/sec):

Datatype	Operation	fory	protobuf	msgpack
MediaContent	Serialize	2,787,853	2,241,239	2,111,882
MediaContent	Deserialize	2,601,783	1,343,196	1,252,873
MediaContentList	Serialize	588,552	443,250	505,927
MediaContentList	Deserialize	570,945	289,274	279,718

• Aggregate throughput (mean ops/sec across all benchmark cases):
  • fory: 5,278,631
  • protobuf: 2,020,540
  • msgpack: 1,982,782