feat: add L2-bypass cache policy to tload

[zhangstevenunity]

Summary

• add an optional typed cache_policy attribute to pto.tload
• preserve legacy TLOAD(dst, src) lowering for absent/default policy
• lower l2_bypass to TLOAD<pto::LoadCachePolicy::L2Bypass>(dst, src) without emitting a target address-alias constant
• expose the policy through C and Python bindings, reject it on A5, and document the A2/A3 behavior
• add lit, Python binding, and PTO bytecode round-trip coverage

Validation

• PTOAS changed-code compliance: 0 errors, 0 warnings
• compiled the changed PTO IR, EmitC lowering, conversion, C API, and Python binding translation units
• generated Python TLoadOp with the cache_policy keyword
• passed ptobc_tload_cache_policy_v0_encode CTest
• passed Python syntax, Bash syntax, and Git whitespace checks

Target dependency

The generated bypass call establishes the PTOAS-to-PTO-ISA contract. PTO-ISA still needs to provide the matching pto::LoadCachePolicy and templated TLOAD implementation before bypass kernels can compile and execute end to end.

Relates to #1356

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.