feat: show validation value even on clean PRs (#6)

Replace the old "Check | Result | Details" table with a new
"What We Checked | How | Result" validation table that communicates
what was validated and the technology used (DataFusion, AST analysis,
pattern scan, column classification).

Key changes:
- New `ValidationSummary` and `QueryProfile` types in `types.ts`
- `extractValidationSummary()` in `cli-check.ts` with `CATEGORY_META`
  mapping for all 7 check categories
- New `query-profile.ts` module for regex-based SQL structure extraction
  (complexity, JOINs, CTEs, subqueries, window functions, aggregation)
- Rewritten executive line: "validated" instead of "modified",
  "N downstream safe" instead of "all checks passed", "N findings"
  instead of separate critical/warning counts
- Collapsible Query Profile section with per-file metadata table
- Footer now includes "Validated without hitting your warehouse"
- Always post comment when `filesAnalyzed > 0` (previously skipped
  clean PRs, which showed zero value)
- `runCheckCommand` now returns `CheckCommandResult` with both issues
  and validation summary
- Version bumped to 0.4.0
- Updated all unit and e2e tests for new format

Co-authored-by: Claude Opus 4.6 (1M context) <[email protected]>

fix: install from npm only, clear notice when check cmd unavailable

- Removed GitHub releases API version fetch entirely
- Install via npm @latest only (GitHub releases are not npm)
- Clear notice when check command isn't available (needs v0.5.10+)
- v0.5.10 must be published to npm to enable AST-based analysis

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

feat: beautified Mermaid DAG — filter tests, visible by default, legend

- Filter out dbt test nodes (not_null_*, unique_*, accepted_values_*)
  from the Mermaid diagram — they clutter the graph
- DAG is now VISIBLE by default (not collapsed in <details>)
  — it's the visual differentiator, show it proudly
- Added color legend below the diagram
- Improved Mermaid styling (thicker strokes, better contrast)
- Node labels displayed using ["label"] syntax for readability
- Shows "(N tests affected)" count in header for filtered tests

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

feat: v0.2.0 — redesigned PR comments, inline reviews, interactive co…

…mmands, 19 rules

PR Comment Redesign:
- Compact summary table (SQL, dbt, Cost, PII in one view)
- Collapsible severity-grouped issues (Critical > Error > Warning > Info)
- ASCII DAG visualization for impact analysis
- Cost before/after/delta per model
- ~8 visible lines for clean PRs, ~12 for complex ones

Inline Review Comments:
- Critical issues always posted on diff lines
- Warnings posted if <= 5 total
- Capped at 10 per PR, batched as single review

Interactive Commands:
- `/altimate review` — full review or specific file
- `/altimate impact` — DAG impact analysis
- `/altimate cost` — cost estimation
- `/altimate help` — command reference

Rule Engine v2 (19 rules):
- 5 new rules: function_on_indexed_column, not_in_with_nulls,
  distinct_masking_bad_join, count_for_existence, no_limit_on_delete
- All rules now have fix suggestions
- Rules categorized: correctness, performance, style, security

329 tests, 0 failures.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

docs: add PR comment design spec from competitive analysis

Based on real competitor PR comments (CodeRabbit, Claude Code Action,
SonarCloud, Snyk, SQLFluff). Includes:
- Detailed analysis of 10+ real PR comments on public repos
- Design principles extracted across all competitors
- 3 full markdown templates (clean, warnings, critical+impact+cost)
- Length guidelines and inline comment strategy

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>