Feat/triton false positive handling

[tsrikris]

Summary

• resolve_triton_source now marks a .py launcher patchable only when the file defines at least one @triton.jit/@gluon.jit kernel. A native or precompiled kernel dispatched through a
  triton-named .py wrapper (zero jit defs) returns unresolved, and resolve_kernel_source falls back to native classification instead of routing it as editable Triton.
• A non-.py launcher sentinel (e.g. "AITER (vendor)") is no longer misclassified as a generated-Triton source. It drives the symbol-name search, so a genuine @triton.jit kernel whose
  trace records only a vendor sentinel launcher resolves to its real source.
• The symbol-index text pre-filter now admits gluon-only files.

Why

Consumers forward the trace's kernel launcher straight through and rely on this resolver as the single oracle for the native-vs-Triton decision. The previous behavior marked any editable .py
launcher patchable without confirming it defines the kernel, mislabeling native/precompiled kernels as patchable; and for vendor-fused ops whose trace records only a sentinel launcher, it
dropped genuinely editable Triton kernels to non-patchable.

Test plan

• tests/test_kernel_source_triton.py (22 passed): native dispatcher .py with no jit def -> non-patchable/native; real @triton.jit/@gluon.jit launcher stays patchable with a pinned
  line; non-.py sentinel launcher -> symbol-index resolution; name-only gluon fallback.
• End-to-end against real traces (5 profiling artifacts): native/precompiled dispatchers through .py wrappers de-route correctly; vendor-fused Triton kernels resolve to their editable
  source; genuinr Triton/gluon routing unchanged.

[codecov-commenter]

Codecov Report

❌ Patch coverage is 91.66667% with 2 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
TraceLens/TraceUtils/kernel_source/triton_pin.py	89.47%	2 Missing ⚠️

📢 Thoughts on this report? Let us know!