fix(mining): add tech-term stopwords and sentence-start filter to entity detector

[arnoldwender]

Summary

Fixes #476 — entity_detector.py was flagging common technical terms (e.g. Handler, Service, Node, Client) as entity candidates, and capturing sentence-opening words like One that happen to be capitalized.

Changes

mempalace/entity_detector.py

• Added _TECH_STOPWORDS frozenset (~60 architecture/runtime/doc terms) merged into STOPWORDS via STOPWORDS = STOPWORDS | _TECH_STOPWORDS
• Replaced re.findall loop in extract_candidates() with re.finditer that checks the 50-character window before each match; words at position 0 or immediately following ., !, ?, or \n are skipped (sentence-start filter)
• Added _SENTENCE_END_RE compiled regex constant for the sentence-end check

tests/test_entity_detector.py

• Updated test_extract_candidates_finds_frequent_names and test_detect_entities_with_person_file to place names mid-sentence (after an opener like "Today", "Later", etc.) so the sentence-start filter does not discard them — reflects correct expected behavior
• All 32 tests pass

Test plan

• Run uv run pytest tests/test_entity_detector.py -v — 32/32 passed
• Verify Handler, Service, Node, Client no longer appear in detect_entities() output on a typical codebase README
• Verify One, Two at sentence start are no longer captured

Closes #476