docs: document bind/implementation/match terminology and their layers

[zzylol]

Summary

design.md's §3 layer-spine table never actually defined "bind" vs "implementation" vs "match", and one passing mention (in §5.1 and the §6.0 crate map) still called asap-plan's L3→L4 pass "the bind pass" -- exactly the collision crates/plan/src/lib.rs's own crate doc explicitly renamed to "implementation" to avoid becoming a fourth/fifth sense of an already-overloaded word.

Adds a terminology table right after the layer-spine table (§3) -- before any per-crate detail section depends on the vocabulary -- covering:

• Bind docs: clean up L3 QueryExpr; add schema flow + L4 SketchExpr + system I/O #1 (L2) -- name resolution, asap_l2::binder::Binder
• Implementation (L3→L4, one node) -- boundary::implementation_for
• implement_tree (L3→L4, whole tree) -- bind::implement_tree (module named bind for historical reasons; content is "implementation")
• Match (a different question from Implementation: "does an available one satisfy a required one") -- boundary::Matcher, ships with no default impl by design; ASAPQuery-backend's Capability::is_satisfied_by is the reference downstream instance
• Bind docs: QuerySpec definition + batched-queries DAG-reuse example #2 (L4→L5) -- deployment-specific placement, e.g. control_plane::sketch_algebra::rules::bind_*

Points back to crates/plan/src/lib.rs's own module doc as the authoritative, code-adjacent source rather than duplicating it wholesale -- that's what should stay in sync with the real implementation; this table is a map, not a second source of truth. Also fixes the two stale "bind pass" mentions elsewhere in the doc.

Note for reviewer

While grounding this I noticed design.md's core::optimizer / core::plan code-sketch sections (§ "Layer 4 framework" / "shared traits bridging layers") describe an OptimizerRule/RuleEngine/BindKllOnQuantile-shaped API that doesn't match what's actually shipped in crates/plan today (boundary::implementation_for, bind::implement_tree, cost_model::CostModel) -- looks like aspirational pre-implementation content that was never reconciled with what got built. Left untouched here (separate, larger scope than this PR), but flagging it.

🤖 Generated with Claude Code