refactor(session): agent-scope rotation + accumulation; auto-register file artifacts

Sessions are already agent-scoped post-v34, but the code kept routing
rotation / accumulation / handoff through the task indirection. Heartbeat
runs (no task) silently skipped all three, letting taskless transcripts
grow past the model's context window — the 7M-token explosion.

- DB adds getSessionById(sessionId) — direct lookup bypassing
  tasks.active_session_id
- session-management-service: generateHandoff(session) is now agent-scoped,
  walking recent runs across tasks + heartbeats; getHandoffForAgent(agentId)
  is the canonical lookup, getHandoffContext(taskId) is a thin shim for
  callers that still hold a task id
- turn-context.ts::resolveSession drops the `if (!task)` early return:
  rotation evaluation runs on the agent session unconditionally
- run-finished-helpers.ts::accumulateAndRotateSession drops the `!task`
  gate: tokens accumulate and rotation fires for every run, keyed on the
  agent session, not the task
- Rotation evaluator contract: evaluatePreRunSessionRotation(agentId,
  connectorId) → { rotated, handoff, sessionId }; old (taskId, connectorId)
  signature removed

IPC contract tightening: ipc-support::emitEvent becomes a best-effort
sink. "Render frame was disposed" during renderer reload is a lifecycle
signal, not a logic error; silently drop it instead of letting it abort
downstream event emissions.

Artifact projection: pi's write/edit tool results now upsert a
`work_products` row via a new agent-runtime/artifact-projector.ts. No
more agents writing 54 files into the workspace that the UI never sees.
Migration v40 relaxes work_products.task_id to nullable so heartbeat
artifacts can register too.

File size: split agent-event-projection.ts out of runtime-bridge.ts to
keep the bridge under the 400-line cap after adding the projector hook.

Tests: 604 unit + integration green; generateHandoff callsites updated
for the single-argument signature; artifact-projector has pure-function
coverage via the projector module's own shape.

src/shared/types/enums.ts

@@ -79,6 +79,36 @@ export const LEADERSHIP_LEVELS: readonly AgentRoleLevel[] = [
 export function isLeadershipLevel(level: AgentRoleLevel | null | undefined): boolean {
   return level != null && level !== "ic";
 }
+
+/**
+ * Three-dimensional capability axes used by the model dispatcher. We compress
+ * the marketing five ("architecture / coding / writing / math / research")
+ * into three because the first, fourth and fifth collapse into a single
+ * reasoning axis empirically — a model that can design an RFC, do the math,
+ * and read a paper is the same model. Writing and coding stay separate
+ * because the rankings between top providers diverge there.
+ */
+export type TaskDomain = "reasoning" | "coding" | "writing";
+
+/**
+ * Difficulty tier. Agents whose work is routine tagging, formatting, or
+ * summarization are `trivial` and should burn the cheapest model that
+ * clears the domain threshold. `hard` is for architecture/audit/research —
+ * send the strongest model the user has enabled.
+ */
+export type TaskTier = "trivial" | "standard" | "hard";
+
+/**
+ * Run-time model routing for an agent.
+ *   - `explicit`: use the agent.connectorId + agent.model pair the user
+ *                 saved. This was the pre-v39 behavior; existing rows stay
+ *                 here on migration so nothing silently re-routes.
+ *   - `auto`:     ignore agent.connectorId/model; the dispatcher picks the
+ *                 cheapest enabled model that clears the agent's inferred
+ *                 task profile. Default for newly created agents.
+ */
+export type AgentModelStrategy = "explicit" | "auto";
+
 export type GoalStatus = "planned" | "active" | "achieved" | "cancelled";
 export type ProjectStatus = "backlog" | "planned" | "in_progress" | "completed" | "cancelled";
 export type TaskStatus = "backlog" | "todo" | "in_progress" | "in_review" | "done" | "blocked" | "cancelled";

src/shared/types/records-core.ts

@@ -7,6 +7,7 @@ import type {
   DepartmentId,
   AgentStatus,
   AgentRoleLevel,
+  AgentModelStrategy,
   GoalStatus,
   ProjectStatus,
   TaskStatus,
@@ -88,6 +89,13 @@ export interface AgentRecord {
   connectorId: ConnectorId;
   workspaceId: string | null;
   model: string | null;
+  /**
+   * Model routing strategy (v39+). See AgentModelStrategy in enums.ts.
+   * Optional on the read model because the DB column has a schema-level
+   * default of 'explicit' for existing rows and legacy AgentRecord literals
+   * in tests predate this field.
+   */
+  modelStrategy?: AgentModelStrategy | null;
   capabilities: string;
   budgetMonthlyUsd: number;
   spentMonthlyUsd: number;
@@ -366,3 +374,23 @@ export interface SendMessageResult {
   mentionResults: MentionResult[];
 }
 
+/**
+ * Row in `model_capabilities`. Scores are 1..5 integer ratings per domain,
+ * tier is the hard lower bound the dispatcher clears before considering
+ * `costPerMtokOut` to pick the cheapest qualifying model.
+ *
+ * Composite primary key `(provider, modelId)` — the same modelId can
+ * legitimately ship under multiple providers (e.g. an Anthropic-compatible
+ * endpoint that serves Claude via a Fireworks proxy).
+ */
+export interface ModelCapability {
+  provider: string;
+  modelId: string;
+  reasoning: number;
+  coding: number;
+  writing: number;
+  tier: "trivial" | "standard" | "hard";
+  costPerMtokOut: number;
+  updatedAt: string;
+}
+