fix: ignore search punctuation in graph queries

[azizur100389]

Summary:

• Normalize graph query search tokens by stripping punctuation before scoring or node lookup.
• Add regressions for punctuated query, path/explain lookup helpers, and CLI query behavior.

Why:

• Fixes graphify query: punctuation in search terms breaks node matching #978, where searches like "extract?" failed to match nodes labeled "extract".

Tests:

• PASS: python -m pytest tests\test_serve.py -k "punctuation or query_terms_strips or query_graph_text_ignores"
• FAIL: python -m pytest tests\test_query_cli.py -k punctuation (local NetworkX does not support the existing test fixture call node_link_data(..., edges="links"); fixture left unchanged per review)
• FAIL: python -m pytest (collection stops because local environment is missing optional dependency: ModuleNotFoundError: No module named 'datasketch' in tests/test_dedup.py)
• PASS: python -m graphify update . (graph rebuilt; graph.html skipped because graph has 5378 nodes, above the default 5000-node visualization limit)

Scope:

• Intentionally avoids unrelated refactors, dependency changes, and generated-file changes.

[safishamsi]

Thanks for the PR! The core fix looks good, but the test file changes are unrelated to the PR's purpose (you've edited test_build.py to modify node_link_data tests that are unrelated to this PR's scope). Could you revert those test changes and keep only the changes relevant to this PR? That will make it easier to review and merge.

[azizur100389]

Addressed, thanks. I reverted the unrelated node_link_data fixture compatibility edits and kept the PR scoped to punctuation normalization plus the focused regression tests. I also updated the PR test section to note that the local CLI regression is blocked by the existing NetworkX fixture call after reverting that change.

[safishamsi]

Implemented in v8 (commit 80301a0). Added _search_tokens helper using re.findall(r'\\w+', ...) that strips punctuation before matching. Updated _query_terms, _score_nodes, and _find_node to use it. Non-Chinese Unicode scripts (Hangul, kana) are handled correctly since the token stripping avoids NFKD normalization on the stored terms. Three new tests (term stripping, _find_node punctuation, query-graph-text integration). All 1390 tests pass. Thanks!