refactor(config)!: derive enable_login_flow from mode, remove ENABLE_LOGIN_FLOW env var

Once OAUTH_SINGLE_AUDIENCE was renamed to LOGIN_FLOW and the validation
gate ensured the only meaningful configuration was
`MCP_DEPLOYMENT_MODE=login_flow + ENABLE_LOGIN_FLOW=true`, the two
controls became redundant. Setting the mode is sufficient; the
ENABLE_LOGIN_FLOW env var doesn't add information.

This commit makes the deployment mode the single source of truth for
the Login Flow v2 toggle:

- `nextcloud_mcp_server/config.py`: drop the `ENABLE_LOGIN_FLOW`
  dynaconf env-var alias. The `enable_login_flow` field stays as an
  internal attribute so the 6 runtime call sites (app.py x4,
  context.py, auth/scope_authorization.py) keep working unchanged.
  Updated field docstring to flag it as derived.
- `nextcloud_mcp_server/config_validators.py`:
  - Drop `enable_login_flow` from `MODE_REQUIREMENTS[LOGIN_FLOW].required`.
  - Drop the validation gate that required ENABLE_LOGIN_FLOW=true for
    LOGIN_FLOW mode (no longer possible to misconfigure — the flag is
    derived, not user input).
  - Add `_sync_derived_flags()` helper called at every return path of
    `detect_auth_mode` to set `settings.enable_login_flow` from the
    resolved mode.
- `tests/unit/test_config_validators.py`: drop `enable_login_flow=True`
  from happy-path fixtures (no longer needed — detection sets it).
  Repurpose `test_login_flow_requires_enable_login_flow_flag` into
  `test_login_flow_mode_auto_derives_enable_login_flow_flag` which
  asserts the new auto-derivation behaviour for both LOGIN_FLOW and a
  non-LOGIN_FLOW mode.
- `docker-compose.yml`: remove `ENABLE_LOGIN_FLOW=true` from the
  `mcp-login-flow` and `mcp-keycloak` profiles.
- `env.sample`: remove the ENABLE_LOGIN_FLOW reference; the comment
  on `MCP_DEPLOYMENT_MODE` now notes the derived flag.
- `docs/configuration.md`, `docs/authentication.md`,
  `docs/login-flow-v2.md`, `docs/auth-flows.md`,
  `docs/troubleshooting.md`, `docs/ADR-025-*.md`: replace
  ENABLE_LOGIN_FLOW=true examples and references with
  MCP_DEPLOYMENT_MODE=login_flow.

BREAKING CHANGE: `ENABLE_LOGIN_FLOW` is no longer read from the
environment. Anyone who relied on `ENABLE_LOGIN_FLOW=true` to activate
Login Flow v2 should set `MCP_DEPLOYMENT_MODE=login_flow` instead (or
rely on it being the default when no other auth env vars are set).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docs/login-flow-v2.md

@@ -58,8 +58,8 @@ NEXTCLOUD_HOST=https://your.nextcloud.example.com
 NEXTCLOUD_OIDC_CLIENT_ID=<your-client-id>
 NEXTCLOUD_OIDC_CLIENT_SECRET=<your-client-secret>
 
-# Enable Login Flow v2 (per-user Nextcloud app-password provisioning for the data leg)
-ENABLE_LOGIN_FLOW=true
+# Select Login Flow v2 mode (per-user Nextcloud app-password provisioning for the data leg)
+MCP_DEPLOYMENT_MODE=login_flow
 
 # App-password storage (required for persistence across restarts)
 TOKEN_STORAGE_DB=/app/data/tokens.db
@@ -172,7 +172,7 @@ mcp-login-flow:
     - NEXTCLOUD_HOST=http://app:80
     - NEXTCLOUD_MCP_SERVER_URL=http://localhost:8004
     - NEXTCLOUD_PUBLIC_ISSUER_URL=http://localhost:8080
-    - ENABLE_LOGIN_FLOW=true
+    - MCP_DEPLOYMENT_MODE=login_flow
     # Dev-only inline value. In production, mount via Docker secret and read
     # from a *_FILE env var or a secrets-management init step.
     - TOKEN_ENCRYPTION_KEY=<your-fernet-key>
@@ -351,7 +351,7 @@ The user revoked it from **Settings → Security → Devices & Sessions**. Delet
 
 ### Multiple worker processes
 
-The provisioning session store is in-memory; `ENABLE_LOGIN_FLOW=true` assumes a single worker. Running with `uvicorn --workers N` will cause provisioning sessions to randomly fail. For higher concurrency, scale horizontally (multiple containers behind a sticky-session load balancer) rather than within a single process.
+The provisioning session store is in-memory; `MCP_DEPLOYMENT_MODE=login_flow` assumes a single worker. Running with `uvicorn --workers N` will cause provisioning sessions to randomly fail. For higher concurrency, scale horizontally (multiple containers behind a sticky-session load balancer) rather than within a single process.
 
 > **Sticky-session keying:** route on the **user identity** (e.g. the `sub` claim from the OAuth Bearer token) — **not** the raw token value, and **not** source IP. Bearer tokens rotate on refresh, which would silently break token-value affinity if a refresh lands between the request that initiates provisioning and the polling request that completes it. MCP clients may also not maintain stable IPs across those requests. A stable per-user identifier extracted from the `Authorization` header (e.g. `sub`) is the right key.