fix: eliminate redundant DB queries in get_learning_paths

[cursor]

Problem

get_learning_paths called get_pattern_sharing as a black box but only used exemplar_sessions from the result. This duplicated four DB queries already performed by get_learning_paths (sessions, engineer names, facets, tool usages) and additionally computed patterns, bright spots, workflow profiles, and model costs — all discarded except the exemplar list. This effectively doubled the database load for the /learning-paths endpoint.

Solution

Extracted the pattern-building and exemplar-derivation logic from get_pattern_sharing into a shared _build_patterns_and_exemplars helper function. Both get_pattern_sharing and get_learning_paths now call this helper with pre-loaded data.

Changes to get_learning_paths:

• Expanded its existing facets query to include extra columns needed for pattern building (outcome, agent_helpfulness, brief_summary)
• Builds session_map, session_facets, session_tools, session_tool_count from data already being queried
• Runs only 2 additional targeted queries (workflow profiles, model costs) that weren't previously available
• Calls _build_patterns_and_exemplars directly instead of get_pattern_sharing

This eliminates 4 redundant DB queries (sessions, engineer names, facets, tool usages) and avoids all the discarded intermediate computation (pattern clustering, bright spots).

Testing

All 21 tests in tests/test_growth.py pass (learning paths, pattern sharing, and onboarding acceleration). Ruff lint and format checks pass.