feat: configurable timeout_keep_alive for keep-alive idle connections

[abhisrao05]

Problem

Flagsmith SDK clients (default poll interval: 10s) intermittently see ConnectionResetError on the pooled /api/v1/environment-document/ request:

urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='10.12.40.89', port=80):
Max retries exceeded with url: /api/v1/environment-document/
(Caused by ProtocolError('Connection aborted.', ConnectionResetError(104, 'Connection reset by peer')))

Root cause

uvicorn's timeout_keep_alive defaults to 5s. Any HTTP intermediary in front of edge-proxy — load balancer, proxy, or the SDK's own urllib3 connection pool that holds an idle connection longer than 5s ends up reusing a half-closed socket on its next request, surfacing as connection reset. Polling SDKs are especially exposed: a 10s poll interval against a 5s server keep-alive guarantees almost every client's pooled connection is stale.

Fix

Surface timeout_keep_alive through edge-proxy's existing config (config.json / env), so operators can raise it past their client poll interval (or LB idle timeout). Default unchanged at 5s — purely additive.

Changes

• src/edge_proxy/settings.py: new timeout_keep_alive: int = 5 on ServerSettings
• src/edge_proxy/main.py: pass timeout_keep_alive=settings.server.timeout_keep_alive to uvicorn.run()

Test plan

• Unit: setting loads from config.json (`{"server": {"timeout_keep_alive": 2}}` → `timeout_keep_alive=2` on the AppConfig instance)
• Integration: with `timeout_keep_alive=2`, server closes idle keep-alive connections at t≈2.08s
• Integration: with `timeout_keep_alive=20`, server closes at t≈20.09s
• Default behavior unchanged when not overridden in config