feat(ir): add a query-DAG JSON exporter + interactive viewer

[zzylol]

Summary

Closes #133 (a query DAG visualization for manual code review/debugging, common query pattern analysis, and — later — CSE analysis, in the spirit of Hamna's bgp-query-dag-explorer).

• crates/ir/src/dag_export.rs — flattens QueryExpr into a generic node/edge graph (exhaustive match over every variant), with a bottom-up structural hash per node so identical subtrees across queries can be spotted cheaply.
• crates/lower/examples/dag_export.rs — lowers one or more --sql/--promql queries and prints a WorkloadGraph as JSON (cargo run -p asap-lower --example dag_export -- --sql "..." --name q1 > dag.json).
• tools/dag-viewer/ — a self-contained, offline HTML viewer (vendored Cytoscape.js + dagre, MIT-licensed, no CDN) with drag-and-drop JSON loading, per-query tabs, a node-detail inspector, and shared-subtree highlighting.

QueryExpr already fully derived Serialize, so the exporter is a thin flattening layer, not a new data model. serde_json moved from a dev-dependency to a real dependency of asap-ir since the exporter needs it at runtime.

Explicitly out of scope (see doc comments / the viewer's README): the shared-subtree highlighting is a client-side structural-hash proxy, not real CSE — asap_plan::cse::dedupe_subtrees/CseWorkloadPlan isn't wired into any end-to-end multi-root planning pipeline today (no caller outside its own unit tests). Wiring that up is a separate, larger effort worth its own issue; once it lands, this viewer is a natural place to render its bindings/Ref output as real converging DAG edges instead of this proxy.

Test plan

• cargo test -p asap-ir — new dag_export unit tests (leaf, chain, n-ary Merge, hash-equality/inequality across identical/differing subtrees) pass; full existing suite still green (41 passed).
• cargo build/test/clippy --workspace and cargo fmt --check all clean.
• Drove the viewer end-to-end in a real Chromium instance (Playwright): loaded a 3-query export, confirmed tab switching, node-click → detail panel, and that the shared Scan node between two queries is correctly outlined and cross-referenced.

🤖 Generated with Claude Code

[zzylol]

@milindsrivastava1997 FYI. I haven't checked how it works lol

[zzylol]

Rebased on current `main` and fixed a compile break: `Aggregate::by` (`GroupKeys`) was renamed to `Aggregate::reduction` (`Reduction`) in #165/#166 after this branch forked, so `dag_export.rs`'s exhaustive match no longer compiled. Updated the `build()` arm and the `chain_preserves_shape_and_child_links` test fixture accordingly (JSON detail key: `"by"` → `"reduction"`).

Verified after the fix: `cargo build --workspace`, `cargo test -p asap-ir` (all 5 `dag_export` tests pass), `cargo clippy -p asap-ir -p asap-lower --all-targets` clean, and a real `cargo run -p asap-lower --example dag_export -- --promql "topk(5, rate(http_requests_total[5m]))"` run producing correct JSON. No other gaps found — the rest of the match is exhaustive against the current `QueryExpr`, and the viewer renders `detail` generically so it needed no changes.

🤖 Generated with Claude Code

[zzylol]

Closing as superseded — this content is already on `main` (landed via #177 after #176's squash-merge initially dropped it due to a stale-cache issue on GitHub's merge endpoint; #177 restored it, see that PR for the full story). Verified: `crates/ir/src/dag_export.rs`, `crates/lower/examples/dag_export.rs`, and `tools/dag-viewer/` are all present and building on current `main`, including the `Aggregate::by` → `reduction` fix from this branch. The only remaining diff between this branch and `main` is that `main` additionally has the README pointing at `dag_export`/`tools/dag-viewer` (from #176/#177) — merging this branch as-is would actually remove that doc section, so closing rather than merging.

🤖 Generated with Claude Code