terminology: the HTS client hangs for minutes behind an egress proxy instead of failing fast

[angela-helios]

Surfaced on PR #466's CI: terminology_integration::test_in_modifier_fails_open_on_hts_unavailable points :in at a closed local port and expects fail-open within the client's 10s timeout — the run took 789s and returned 500.

Three compounding problems in TerminologyServiceClient::new:

1. reqwest honors the runner's proxy env but cannot parse CIDR entries in no_proxy, so 127.0.0.1 traffic detoured through the egress proxy (the ci: nightly ui-tests-matrix — Elasticsearch leg failing (Postgres and MongoDB green) #449 runner-proxy signature, third surface).
2. No connect timeout — an unreachable server consumed the whole request budget instead of failing in milliseconds.
3. build().unwrap_or_default() silently fell back to a client with no timeouts at all when the builder errs — the unbounded hang observed.

The terminology server is a configured internal endpoint; the client should bypass proxies outright, fail connects fast, and treat a client-build failure as the startup-visible misconfiguration it is.