fix: Check disabled state before AI Config mode mismatch

[jsonbailey]

This PR fixes two related issues in LDAIClientImpl._evaluate() in the server-ai package, both around how a served AI Config variation's mode is reconciled with the mode the caller requested.

Fix 1 — Check disabled state before the mode comparison

Bug: A disabled AI Config variation is served by LaunchDarkly as {"_ldMeta": {"enabled": false}} with no mode field. The mode-mismatch check ran before the enabled check, so the absent mode defaulted to 'completion'. Calling agentConfig() (or judgeConfig()) on a disabled config then made 'completion' !== 'agent' true and emitted a spurious warning:

AI Config mode mismatch for sample-agent: expected agent, got completion. ...

The config was simply disabled, not a genuine mode mismatch.

Fix: Short-circuit disabled variations before the mode comparison. When value._ldMeta?.enabled === false, _evaluate() returns LDAIConfigUtils.createDisabledConfig(key, mode, ...) of the requested mode — no mode comparison, no warning. (=== false matches the existing disabled-check convention in agentGraph().)

Fix 2 — Return the caller's default on a genuine mode mismatch

Bug (pre-existing): When a config is enabled but its mode genuinely differs from what the caller requested (e.g. an enabled completion config requested via agentConfig()), js-core logged "...Returning disabled config." and returned an empty disabled config. Per the AI Config spec (the "mode mismatch returns the caller's supplied default" rule, merged to sdk-specs main via PR #229 / AIC-2207), the SDK must instead return the caller's supplied default of the requested mode. .NET already does this; js-core was off-spec.

Fix: In the genuine-mismatch branch (which now only runs for enabled configs, after Fix 1's disabled short-circuit), build the returned config from the caller's supplied default (converted to the requested mode via the already-computed ldFlagValue) instead of an empty disabled config. The warning text now reads "Returning caller's default." to match the new behavior and .NET's wording. Matching-mode and disabled paths are unchanged.

Tests

• Replaced the old "returns disabled config" mismatch test with one asserting the returned config equals the caller's default (enabled, model, provider, messages) and that the warning is still emitted.
• Disabled-variation regression tests (agent / judge / completion served as {"_ldMeta": {"enabled": false}}) assert no mode-mismatch warning and a disabled config of the requested type — unchanged and still passing.
• All 225 server-ai tests pass; lint clean.

Companion changes

• Spec change in sdk-specs on branch jb/disabled-config-mode-check.
• Companion SDK fix in dotnet-core on branch jb/disabled-config-mode-check.

No changelog or version edits (handled at release time).

BEGIN_COMMIT_OVERRIDE
fix: Check disabled state before AI Config mode mismatch
fix: Return caller's default on AI Config mode mismatch
END_COMMIT_OVERRIDE

[github-actions]

@launchdarkly/js-client-sdk size report
This is the brotli compressed size of the ESM build.
Compressed size: 31978 bytes
Compressed size limit: 34000
Uncompressed size: 114243 bytes

[github-actions]

@launchdarkly/js-sdk-common size report
This is the brotli compressed size of the ESM build.
Compressed size: 26365 bytes
Compressed size limit: 29000
Uncompressed size: 129044 bytes

[github-actions]

@launchdarkly/js-client-sdk-common size report
This is the brotli compressed size of the ESM build.
Compressed size: 38739 bytes
Compressed size limit: 39000
Uncompressed size: 212244 bytes

[github-actions]

@launchdarkly/browser size report
This is the brotli compressed size of the ESM build.
Compressed size: 179579 bytes
Compressed size limit: 200000
Uncompressed size: 831422 bytes