boost exact-match seeds and stop hubs hijacking query output

[ReverendJack3000]

What

_score_nodes gets an EXACT_MATCH_BONUS for terms that equal a node's norm_label (or norm_label with trailing () stripped, since the AST extractor emits function labels as foo()), and _subgraph_to_text accepts an optional seeds arg so the rendered output emits seeds first in score order before the BFS expansion sorted by degree desc.

Why

Single-token identifier queries — by far the most common usage — get buried under high-degree hub modules. Two issues compound:

1. _score_nodes only used substring matching, so the seed function and every sibling that mentioned the symbol tied at score 1; ties broke alphabetically on node id.
2. _subgraph_to_text always re-sorted the rendered nodes by G.degree(n) descending, so even when _score_nodes returned the right seed, the hub it expanded into (e.g. app.js, controller.js) always rendered first.

Before

$ graphify query "pasteFromClipboard" --budget 600
NODE app.js [src=frontend/app.js loc=L1 community=3]
NODE now() [src=frontend/modules/core/perf.js loc=L1 community=1]
NODE .filter() [src=backend/tests/_fake_inventory.py loc=L80 community=6]
NODE draw() [src=frontend/app.js loc=L4047 community=3]
NODE FakeInventoryQuery [src=backend/tests/_fake_inventory.py loc=L18 community=1]
...

pasteFromClipboard() does not appear in the first 30 NODE lines on a ~6k-node graph.

After

$ graphify query "pasteFromClipboard" --budget 600
NODE pasteFromClipboard() [src=frontend/app.js loc=L4847 community=3]
NODE app.js [src=frontend/app.js loc=L1 community=3]
NODE now() [src=frontend/modules/core/perf.js loc=L1 community=1]
...

Scope of behavior change

• seeds=None in _subgraph_to_text preserves the previous degree-desc ordering exactly. The only call sites updated to pass seeds= are _tool_query_graph (MCP) and the query CLI handler in __main__.py. _tool_shortest_path and the path CLI use _score_nodes only for endpoint resolution and do not render via _subgraph_to_text, so they automatically benefit from the bonus without further changes.
• Multi-term / prose queries are unaffected because the bonus only fires on term == norm_label exactly. Substring scoring is untouched.
• EXACT_MATCH_BONUS = 100.0 swamps any plausible substring sum (substring score caps at ~1.5/term).

Tests

Six new cases in tests/test_serve.py:

• test_score_nodes_exact_match_beats_substring — exact match outranks substring distractors on the same graph.
• test_score_nodes_exact_match_strips_function_parens — saveDiagram() matches term savediagram.
• test_score_nodes_exact_match_no_false_positive — unrelated term still returns no matches.
• test_subgraph_to_text_seeds_render_first — seed before hub when seeds=[...].
• test_subgraph_to_text_no_seeds_preserves_legacy_order — seeds=None matches no-arg call (back-compat guard) and hub still wins on degree.
• test_query_pipeline_exact_match_ranks_above_hub — end-to-end _score_nodes -> _bfs -> _subgraph_to_text(seeds=...).

Full suite (447 tests) green.

[safishamsi]

The requested changes have been implemented in 3fdae8f. Added _EXACT_MATCH_BONUS = 100.0 to _score_nodes for exact label matches (strips trailing () for functions), and seeds now render first in _subgraph_to_text before the degree-sorted expansion. Shipped in v0.6.2.