Add standalone per-case golden scripts for 4.5-4.9 samples

[HecreReed]

Summary

• add standalone per-case *_golden.py / *_compare.py under test/samples/* for the PTO manual 4.5–4.9 cases
• make generate_testcase.py prefer sample-local custom assets and emit validation_meta.env
• make remote validation honor custom golden/compare scripts instead of self-copying NPU outputs as golden
• sync sample-local custom assets into the CI payload

Validation

• python3 -m py_compile test/npu_validation/scripts/generate_testcase.py test/samples/*/*_golden.py test/samples/*/*_compare.py
• smoke: standalone golden.py + compare.py for VectorAddition/vectorAddition and Cmp/cmp
• integration smoke: generate_testcase.py with custom asset pickup for representative cases in local temp dirs
• remote subset via test/npu_validation/scripts/run_remote_npu_validation.sh using handcrafted payload:
  • pass: shls, vectorAddition, sel, expands
  • diff remaining: colsum, cmp

Notes

• this PR is Draft because colsum and cmp still show golden diff in the remote subset run and need follow-up

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 9643c7cee3

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Compute pure column reduction in colsum golden

The custom colsum oracle currently adds the reduced sum to an initialized destination buffer (reduced + out), but pto.tcolsum semantics are dst[0,j] = sum_i src[i,j] with tmp only as workspace (see docs/PTO_IR_manual.md), so this produces an incorrect golden for this case and causes deterministic compare failures when validating colsum.

Useful? React with 👍 / 👎.