fix(rust): deep clone TypeMeta to prevent UB in concurrent scenarios

[BaldDemian]

Why?

As mentioned in #3490 (comment), one cargo bench case stably fails in the current code. After some digging, I think it results from this line:

fory/rust/fory-core/src/resolver/type_resolver.rs

Line 1389 in ede9b64

type_meta_by_index: self.type_meta_by_index.clone(),

The clone() method for TypeResolver should return a deep clone:

fory/rust/fory-core/src/resolver/type_resolver.rs

Lines 1316 to 1319 in ede9b64

	/// # Returns
	///
	/// A deep clone of the TypeResolver with all internal Rc instances recreated.
	/// This ensures thread safety when cloning from multiple threads simultaneously.

But the type_meta_by_index field has type Vec<Option<Rc<TypeMeta>>>, calling clone() on this type directly would lead to a shadow clone indeed.

The internal_type_info_by_id field in TypeResolver also has a similar type Vec<Option<Rc<TypeInfo>>> but it doesn't call clone() directly.

What does this PR do?

• deep clone type_meta_by_index. Now the cargo bench case can run successfully and stably.

• introduce bazel-setup GitHub action in C++ sanitizer job

• the existing test cases in test_multi_thread.rs are a little too weak to expose tricky concurrent issues. I wrote a new test case copying the main logic of the failed cargo bench case. Without the fix, this new test case would fail with a high chance.

Related issues

No.

AI Contribution Checklist

No.

Does this PR introduce any user-facing change?

No.

Benchmark

No.

[chaokunyang]

Looks great!