topics: add pr

[khezen]

Changed

• Renamed experiences section to actions in ContextMemory (section prefix "ex" → "ac")
• max_context_item_tokens default: 410 → 512 (capacity: ~32 items with 16384 context memory tokens)
• Hippocampus init parameter topic_ids: list[str] → topics: list[Topic] — topics auto-registered in context memory
• Summarizer.forward() signature simplified: individual PR fields replaced with pr_context: PRContext parameter
• Scope resolver no longer builds dependency graphs between scopes
• Per-signature iteration defaults: doc (max_iters=1, max_llm_calls=2), scope (max_iters=3, max_llm_calls=5), code_review (max_iters=5, max_llm_calls=8)

Added

• PRContext.pr_url and PRContext.pr_description fields
• PRContext.to_topic() helper — builds a Topic from PR metadata for cross-review memory
• PR URL topic automatically registered across all pipeline modules (scope resolver, summarizer, code reviewer, doc reviewer, supply chain auditor, auditor)
• GitHub Action inputs: doc-max-iters, doc-max-llm-calls

Removed

• Topic.dependencies field and all dependency resolution logic in scope resolver
• ScopeResult.is_dependency field
• Dependency extraction from manifest_parser.py (~578 lines): PACKAGE_MANAGER_TO_ECOSYSTEM, extract_dependencies(), infer_repo_from_name(), and 13 per-language _extract_deps_from_* functions — only extract_package_name() retained