refactor: make evaluator cast dispatch static

[KKould]

What problem does this PR solve?

Evaluator casts were still selected through runtime type matching, which added avoidable overhead and kept a lot of cast-related logic and tests concentrated in dispatcher-style code.

There were also several no-op cast paths that still flowed through binding/execution even when the source and target types were already identical, plus the projection executor was still rebuilding its temporary arena values buffer for each row.

Issue link:

What is changed and how it works?

• Make evaluator cast dispatch static so the concrete cast evaluator is chosen up front, and keep eval_cast as the only cast execution entrypoint.
• Remove the old dynamic cast-selection flow and update binder/expression paths to rely on the bound cast evaluator instead of scattered runtime type checks.
• Short-circuit same-type casts during binding so redundant cast expressions can be skipped early.
• Push Cow<LogicalType> through the relevant evaluator factory / type-combination paths to reduce unnecessary cloning where the type only needs to be borrowed.
• Split evaluator dispatch code into dedicated binary.rs, unary.rs, and cast.rs files while keeping per-type implementations in the existing flat evaluator files.
• Move semantic cast/binary tests next to the corresponding type implementation files, and expand cast coverage so each evaluator file tests the target types it actually implements.
• Reuse the projection executor arena temporary values buffer across rows instead of rebuilding it each time.

Code changes

• Has Rust code change
• Has CI related scripts change

Check List

Tests

• Unit test
• Integration test
• Manual test (add detailed scripts or steps below)
• No code

Side effects

• Performance regression: Consumes more CPU
• Performance regression: Consumes more Memory
• Breaking backward compatibility

Note for reviewer

The main review surface is:

1. static cast dispatch and same-type cast short-circuiting across binder/expression/evaluator creation
2. evaluator module reshaping plus the per-type cast test relocation/coverage expansion
3. projection executor temporary buffer reuse

Validation used for this branch:

• cargo test -q