Doc trees phase 1: trees store + explicit `about` + treeId on documents

[ddisisto]

Internal refactor that lays the data-model foundation for user-created doc trees (#8), with an opportunistic attention-store index added while the schema is already being recreated. Deliberately ships no visible change — the doc-list panel looks and behaves identically — but introduces the primitives that #11, #12, #13 will build on, and kills the quadratic scaling in getDocumentAttention before any real data accumulates.

Context

Project is at single-user PoC stage with no external data to preserve. This issue uses the freedom to ship the simplest possible schema change rather than carrying back-fill logic across the version bump. Since onupgradeneeded(v2) is already recreating the attention store, a bySession index is added in the same pass — the cheapest possible moment to fix the existing full-table scan in getDocumentAttention and getLastViewedBlock.

Decisions this issue bakes in

(From the planning discussion on #8.)

• Coexistence over uniformity. Bundled docs keep path-shaped IDs (docs/architecture.md); user-tree docs will get opaque IDs in Doc trees phase 3: import via typed entry and clipboard paste #12. StoredDocument.treeId is the discriminator. The two ID shapes reflect a real asymmetry (filesystem-structured vs. not), not tech debt.
• Explicit about. The about tree becomes a real row in IDB, not the hardcoded label in doc-list.ts:154. System trees are distinguished from user trees via kind.
• Clean-slate upgrade. onupgradeneeded(v2) drops existing stores and recreates them fresh. No back-fill, no conditional open-path logic. On next open, the bundle reimports everything from scratch.
• Attention index in the same schema bump. Add a bySession index on the attention store and rewrite the two query sites to use it. Not a separate issue — folding it in now avoids a follow-up DB_VERSION bump later when there's actual data to worry about preserving.

Changes

store.ts

• Bump DB_VERSION 1 → 2.
• onupgradeneeded(v2):
  • Delete the existing documents, blocks, reading_sessions, attention stores if present.
  • Recreate documents, blocks, reading_sessions with their current schemas.
  • Recreate attention with a new bySession index on sessionId.
  • Create new trees object store, keyPath: 'id'.
  • Seed the canonical about record: { id: 'about', label: 'about', kind: 'system', created: Date.now() }.
• StoredDocument gains treeId: string (strictly non-optional — no legacy rows exist post-upgrade).
• New StoredTree interface:

  interface StoredTree {
    id: string            // 'about' for system, 'u:<uuid>' for user trees
    label: string
    kind: 'system' | 'user'
    created: number
  }

• New tree helpers: listTrees(), getTree(id), putTree(tree), deleteTree(id). deleteTree is unused in this issue but needed for Doc trees phase 2: create/rename/delete empty user trees #11.
• Rewrite getDocumentAttention to use the new bySession index:

  // before: for each session, getAll() the whole attention store (O(sessions × total_records))
  // after:  for each session, index.getAll(session.id) (O(records for that session))

• Rewrite getLastViewedBlock with the same pattern — currently also does a single unscoped getAll() on the attention store (store.ts:334-335).
• Also update deleteDocumentData: the cursor-scan over the whole attention store (store.ts:249-257) can become per-session index.getAllKeys(sessionId) + delete. Same asymptotic improvement, same index, same price.

documents.ts

• importFromBundle stamps treeId: 'about' on every StoredDocument it creates.
• No back-fill logic anywhere (there are no legacy rows).

doc-list.ts

• refresh() reads tree records via listTrees() and groups docs by treeId — the about label stops being hardcoded at line 154.
• For the one tree that exists right now, output is identical to today.
• No change at the call site for the attention query improvements — buildLeaf keeps calling getDocumentAttention(path), the speedup is internal.

Acceptance

• Upgrade from v1 → v2 drops and recreates stores; existing data is wiped (expected and acceptable for the single-user PoC).
• After upgrade, trees store contains exactly one record: the canonical about entry.
• The attention store has a bySession index on sessionId.
• getDocumentAttention returns the same results as before, but via the index — no full-store scan inside the session loop.
• getLastViewedBlock and deleteDocumentData likewise use the index and no longer scan the whole attention store.
• All bundled docs get treeId: 'about' set at import time; no docs have a missing treeId.
• Doc-list panel renders identically to pre-refactor — same tree, same labels, same attention stats, same behavior.
• Fresh install (no prior IDB) produces the same end state as an upgraded install.
• Poll-refresh loop in doc-list.ts no longer reads the full attention store per tick.

Touchpoints

• frontend/src/store.ts — schema bump, new trees store, bySession index on attention, CRUD helpers, three query rewrites
• frontend/src/documents.ts — importFromBundle stamps treeId
• frontend/src/doc-list.ts — refresh/buildTree read from the trees store

Out of scope

• Any visible UI (lands in Doc trees phase 2: create/rename/delete empty user trees #11)
• Any import primitives (Doc trees phase 3: import via typed entry and clipboard paste #12, Doc trees phase 4: import via file upload (drag-drop + picker) #13)
• Link resolution changes (user trees don't exist yet)
• Further query optimisations (e.g. a byDoc composite index on reading_sessions) — only the attention hot path is in scope

Parent: #8