test(config): fix all 9 stale failures in the config suite (#12774)

[mrveiss]

Thinking Path

config/ was 9 failed, 93 passed. Every one turned out to be a test asserting an API, module or value that no longer exists — the same class as #12668 (stale attributes) and #12673 (environment-dependent assertions) — not product bugs. So the work was verifying what each subject actually does now, then retargeting the test at that, rather than "making it pass".

The recurring trap: ConfigManager.get() and .set() are flat dict operations (config/sync_ops.py). They resolve neither dotted paths nor env vars. Six of the nine failures trace back to tests assuming otherwise.

What Changed

validation_test (2) — GH#9232 removed backend.server_host from the required set; _REQUIRED_CONFIG_KEYS is now literally empty because it defaults to 0.0.0.0. The tests asserted it was required.

Rather than delete the coverage, I kept the invariant the class exists for. Its docstring is "Ensure is None (not truthiness) is used for required-key checks" — so that distinction is now exercised against memory.redis.host, which is still conditionally required and is still gated on if value is None. The falsy-but-present side ('0') was already covered.

timeout_configuration_test (1) — assigned config.llm_timeout = "999.0" then asserted "AUTOBOT_LLM_TIMEOUT" in os.environ. Assignment to the proxy doesn't write the environment; the test's own comment conceded it only "test[ed] the mechanism exists".

The real mechanism is pydantic-settings reading env at construction, and the module-level proxy is built once and cached. Verified directly before rewriting:

fresh AutoBotConfig().llm_timeout with env set:  999
fresh without env:                                30

config_migration_integration_test (6) — each verified against the live objects:

Assumption	Reality
cm.get("a.b.c")	flat lookup — needs get_nested()
cm.set("a.b.c", v)	flat write — needs set_nested(); this is why the processor read 0.7 not the 0.9 the test "set"
get_section()	renamed get_config_section()
from utils.config_manager import config	module removed
from config import config	yields SSOT _ConfigProxy — has neither .get nor .config; the legacy wrapper is config.compat.Config
"llm" top-level section	LLM config lives under backend.llm
llm.orchestrator_llm / llm.ollama.port	do not exist anywhere — the old equality assertion compared None to None and was vacuous
redis.host in ["localhost","127.0.0.1"]	an SSOT deployment value — pinned assertion fails on any real deployment

Verification

before (unmodified tree):   9 failed, 93 passed
after:                    102 passed

Two of these were fixed by making the assertion stronger, not weaker: the backward-compat check now compares a key that actually exists (the previous one was vacuously true), and the required-key test now targets a key that is genuinely still required.

Model Used

Claude Opus 5 (1M context)

Closes #12774

[github-actions]

✅ SSOT Configuration Compliance: Passing

🎉 No hardcoded values detected that have SSOT config equivalents!