feat(enhance): the degenerate exits get their own diagnostic channel — exit_kind + the incomplete_summary histogram (#615)

[gadievron]

Fixes #615.

What was wrong

The agentic enhance loop's four degenerate exits are conflated under one classification: "incomplete", but they differ in kind and remedy: three cheap model-behavior exits (end-turn-without-finish, finish truncated at max_tokens, no-tool-calls — confidence 0.3, remedied by retry/prompt work) vs one budget-exhaustion exit (MAX_ITERATIONS — confidence 0.2, remedied by raising the budget, not retrying). In the referenced run the split was 41 "Analysis incomplete" + 4 "Could not complete analysis within iteration limit" — a 4-unit class a consumer reading only classification cannot distinguish from the 41.

The fix

No fourth classification (that would break every == "incomplete" consumer — checkpoint.py, enhancer.py, the degenerate-exit tests). Instead:

• exit_kind on AgentResult — the four exit sites stamp their kinds (end_turn_without_finish / finish_truncated / no_tool_calls / max_iterations); a completed analysis carries the empty string and serializes without the key (present-only — the completed-record byte-identity).
• The incomplete_summary histogram beside the existing error_summary: the enhancer's counting loop aggregates exit_kind (legacy unstamped rows fall back to "unstamped"); EnhanceResult carries the histogram (present-only in to_dict); and both step-report writers — the scanner's pipeline summary and the standalone CLI's enhance summary — thread it, so the 4-vs-41 split is visible in the artifacts without stderr mining.
• The classification and every existing consumer are unchanged (pinned by the classification-unchanged test).

Evidence

• RED (pristine): AgentResult had no field; the four exits stamped nothing; no histogram anywhere.
• GREEN: 8/8 on the new file — the four kinds + the completed negative (present-only), an incomplete serializing its kind, the classification-unchanged pin, the histogram census (max_iterations 2 / end_turn 1 / unstamped 1), the EnhanceResult carry + the present-only negative, both summary writers' pins, and two real-exit receipts driving the actual agent loop through the existing degenerate-exit harness (the end-turn-without-finish and the R2-B finish-truncated stamps on real AgentResults, not hand-constructed ones); the full suite 4171 passed + 33 skipped; ruff clean; Semgrep 0.
• The hunt (one round, all findings fixed): the CLI's enhance summary was a second writer the first pass missed (threaded); the stamp sites were untested against the real loop (the two real-exit receipts added); the typing convention + a dead splat (cleaned). The #635 merge shape named (four adjacent-insert hunks; the post-merge invariant sum(incomplete_summary.values()) == incomplete_count recorded for the reconciler).

Notes for the reviewer

• The four same-anchor EnhanceResult edits will meet feat(analyze): the enhancement non-verdicts are visible, out of the prompt, counted, and migrated on re-enhancement (#611) #635's incomplete_count/total_units edits on merge — keep both; the post-merge invariant to pin: the histogram's sum equals incomplete_count.
• The context_enhancer's internal _summary_incomplete count and the resume path are kind-blind by design (they count the classification, not the kind) — out of this issue's stated scope ("alongside error_summary").

Related: #611 (the sentinel-handling round that surfaced this), #609 (the same stage's metadata-loss fix).