Refine autobot summarization and labeling

Adjust autobot workflows to better handle large PRs and issues: lower per-window thresholds and token budgets, introduce window variants (full/compact/minimal) and batching to build multi-window summary prompts, and add logic to downgrade coverage or create an overflow batch to stay within limits. Add per-batch AI steps (up to 10), aggregate the batch analyses into a single summary prompt, and reduce max completion tokens for the final summary.

Enhance issue triage and labeling: add an issue summary flow, expand the label prompt to accept issue analyses, add label inference and normalization helpers, support new labels (improvement, proposal), and improve label deduplication/selection logic. Introduce major-release alert creation/removal based on breaking-change labels.

Minor: refactor helper functions (topDirectoryForFile, scoring/tiebreakers), write prompt files to /tmp, and update smart-link rules and regex to treat "implement" as a close directive and allow closes for non-bug items.

Author: YaronKoresh