Add scoped overviews, hotspots, and graph freshness warnings

[forhappy]

Large graph overviews could not be narrowed to a source area, dependency hotspots required expensive broad output, and answers gave no notice when their index was stale. This adds source-scoped architecture and community views, compact hotspot rankings, and advisory freshness checks against the selected artifact's recorded repository.

• Add hotspots and MCP get_hotspots, with deterministic distinct-contact and incoming-dependent rankings. Inferred and document relationships are optional labelled layers; unresolved and ambiguous targets remain excluded. Scoped rankings retain dependencies outside the selected area.
• Add repeatable path/module scopes to architecture and a bounded community command. Machine outputs use explicit versioned schemas and disclose omissions.
• Add bounded, offline Git freshness checks to CLI and MCP answers. Preserve the build revision through impact/traversal caches; show recovery commands without changing CLI JSON stdout or immutable history.
• Replace missing Program IR/graph failures with actionable commands and validate Program subcommands before artifact access. Update compatibility, migration, security and command references.

Validation:

• Passed workspace library/binary tests and workspace library/binary Clippy with warnings denied.
• Passed CLI/MCP tests, including product contracts, scoped views, budgets, cold/warm traversal caches, and executing the missing-Program recovery command.
• Passed product-boundary and whitespace checks.
• Final regression rerun and formatting check passed. The full native code-graph fixture gate passed, including semantic/scale assertions, topology checks and release-build React qualification.

Compatibility: unscoped architecture and existing MCP community defaults retain their meaning. Scoped architecture wraps the original result; MCP envelopes can include optional freshness. Disposable traversal caches rebuild automatically. Graph and historical schemas remain unchanged.

Public qualification: independent graph-record checks matched hotspot rankings, scope boundaries and community membership on pinned Cobra, Flask, Gson, Zod and Axum graphs. Default hotspot text used 427–614 estimated tokens. The frozen-graph warm-index replay passed 22/25 Compass questions versus 18/25 Graphify questions. Mean estimated output was 516 versus 758 tokens; on the 16 questions both tools passed, it was 511 versus 756. This is a debug correctness/token replay with a 180-second query cap, not a latency comparison. Two broad Compass questions missed required markers and one Zod path query timed out. The separate cold run and repeated census timeout are retained; no claim of cold-build speed or complete concept recall is made.

Freshness preflight refuses configured conversion filters and submodules before working-file checks; the CLI regression proves a configured clean filter is never executed. Invalid scope errors remain under 200 bytes without terminal controls. All 14 GitHub CI checks passed on final head e3ae229f5c65a1296d1fd7114b827bc7789b9669; the PR is mergeable.