🔧(compose) make the LiveKit agents opt-in in the local dev stack

`make run` no longer starts the metadata collector and the
multi-user transcriber. On my machine, this saves about 400 MB
of RAM and almost 2% CPU, out of the 11% the stack uses.
Start them with `make run-agents` when needed.

Disable `METADATA_COLLECTOR_ENABLED` by default in `common.dist` so
the backend does not dispatch jobs to an agent that is not running,
and document how to enable each agent in `developping_locally.md`.

Makefile

@@ -169,7 +169,7 @@ run-summary: ## start only the summary application and all needed services
 	@$(COMPOSE) up --force-recreate -d celery-summary-summarize
 .PHONY: run-summary
 
-run-agents: ## start the multi-user-transcriber agent
+run-agents: ## start the LiveKit agents (opt-in, see docs/developping_locally.md)
 	@$(MAKE) run-agent-multi-user-transcriber
 	@$(MAKE) run-agent-metadata-collector
 .PHONY: run-agents
@@ -186,7 +186,6 @@ run:
 run: ## start the wsgi (production) and development server
 	@$(MAKE) run-backend
 	@$(MAKE) run-summary
-	@$(MAKE) run-agents
 	@$(COMPOSE) up --force-recreate -d frontend
 .PHONY: run
 

docs/developping_locally.md

@@ -107,6 +107,37 @@ $ npm i
 $ npm run dev
 ```
 
+### LiveKit agents (optional)
+
+The LiveKit agents are not started by `make run`. Each one runs its own
+container and stays connected to LiveKit, which costs CPU and memory you
+don't need unless you work on the features they power. Start them only
+when you need them.
+
+| Agent | Feature | Make command | Setting in `env.d/development/common` |
+|---|---|---|---|
+| `metadata-collector-dev` | Recording metadata (used to identify speakers in transcripts) | `make run-agent-metadata-collector` | `METADATA_COLLECTOR_ENABLED=True` |
+| `multi-user-transcriber-dev` | Live subtitles | `make run-agent-multi-user-transcriber` | `ROOM_SUBTITLE_ENABLED=True` |
+
+To start both at once:
+```shellscript
+$ make run-agents
+```
+
+Then set the matching settings to `True` and restart the backend so it
+picks them up:
+```shellscript
+$ make run-backend
+```
+
+The multi-user transcriber also needs a speech-to-text provider. Configure
+`STT_PROVIDER` and its credentials in
+`env.d/development/multi_user_transcriber`.
+
+Keep the settings and the agents in sync: if a setting is `True` while its
+agent is stopped, the backend still dispatches jobs to it and the feature
+fails silently.
+
 ---
 
 ## Adding Content

env.d/development/common.dist

@@ -104,10 +104,11 @@ ROOM_TELEPHONY_ENABLED=True
 # ROOMKIT_ENABLED = True
 # ROOMKIT_SERVER_TO_SERVER_API_TOKEN = ThisIsAnExampleKeyForDevPurposeOnly
 
-# Metadata
-METADATA_COLLECTOR_ENABLED=True
+# LiveKit agents (opt-in, start them with `make run-agents`)
+# Metadata (requires the metadata-collector agent)
+METADATA_COLLECTOR_ENABLED=False
 
-# Subtitle
+# Subtitle (requires the multi-user-transcriber agent)
 ROOM_SUBTITLE_ENABLED=False
 
 FRONTEND_USE_FRENCH_GOV_FOOTER=False