feat(logging): tag logs with a per-session ID

[aqandrew]

What

Introduce a per-session identifier and make the extension tag every log line
with it, so all logs for a session can be correlated by searching a single ID.

This is Phase 1 of 3 for
DEVEX-661 — "VS Code: add
session_id to all requests, logs, existing telemetry, and CLI invocations".

Changes

• Add SessionLogger (src/logging/sessionLogger.ts) — a Logger that wraps
  the Coder output channel and prefixes every message with [<sessionId>].
• Generate the session ID once in ServiceContainer (reusing the existing
  newSessionId() that already backs the telemetry session, so logs, telemetry,
  requests, and the CLI all share one ID), and expose it via getSessionId()
  for the later phases.

Because every service already receives Logger by injection, session-tagged
logging propagates with no call-site changes.

Design notes

The code owner (Ehab) suggested composing the session ID and the VS Code logger
inside container.ts's ServiceContainer; this PR follows that approach. The
telemetry sessionId from newSessionId() is the RFC session_id (confirmed
with Ehab — reuse it rather than minting a second ID).

Testing

• pnpm typecheck, targeted pnpm lint, and pnpm test:extension (full suite:
  2108 passing).
• New unit tests for SessionLogger prefixing and argument forwarding.

Implementation plan (DEVEX-661)

Thread a single per-session identifier (16-byte / 32-char lowercase hex) so that
logs, API requests, existing telemetry, and the CLI ssh invocation the
extension drives can all be correlated by one session_id.

In-scope RFC requirements

Req	Summary	Where
1	Generate 16-byte / 32-hex session ID	Reuse existing newSessionId()
2	Session ID on every client log	Phase 1 (this PR)
3	session_id on every API request via baggage	Phase 2
4.2	session_id on VS Code telemetry	Phase 2 (already covered by reusing the ID)
5.2/5.3	CODER_TRACE_SESSION_ID via process.env + terminal env collection	Phase 2
7	Log workspace/agent/lifecycle state changes at info	Phase 3
16	Default --log-dir + old-log cleanup	Already implemented — verify only

Out of scope: coderd tracing middleware (req 6), agent/coordination-protocol
changes and the CLI ssh subcommand behavior (reqs 8–15) live in coder/coder.
Req 13 (in-memory log buffer flushed on connection failure) is deferred to a
follow-up.

Key decisions

• D1 (resolved): the telemetry sessionId is the RFC session_id — one ID,
  generated once in ServiceContainer.
• D2: session scope is per activation; Remote-SSH opens each workspace in a fresh
  activation, so container-scoped ≈ per-connection.
• D3 (resolved): prefix every log line with the full 32-hex ID as
  [<sessionId>] <message>.
• D4: baggage: session_id=<hex> header.

Phases

1. Session ID + SessionLogger composition in the container (this PR).
2. Thread the ID to API requests (baggage), telemetry (already covered), and the
   CLI via CODER_TRACE_SESSION_ID on process.env + the terminal env
   collection.
3. Log workspace / agent / lifecycle state transitions at info.

---

🤖 Generated by Coder Agents.

[linear-code]

DEVEX-661

[EhabY]

Should we use shortId for the session ID? This is very verbose now and conflict is extremely unlikely to happen

Also I'm not sure if this is clear that the first number is the session ID while the second number is the request ID

[aqandrew]

Yeah we definitely should use shortId. What do you think of wrapping both the session id and request id in square brackets, and labeling them? So the top line would become

2026-08-13 14:53:01.305 [trace] [session ece5f89f] ← [request 38b0a2b4] 200 GET /api/v2/workspaces?q=owner%3Ame (158 kB) 1.35s