feat: flagd-core in-process evaluator returns reason=DISABLED for disabled flags

[jonathannorris]

Context

Part of open-feature/flagd#1965. ADR: docs/architecture-decisions/disabled-flag-evaluation.md

The flagd-core library (tools/flagd-core) replicates the flagd evaluation logic in-process. Today it returns an error for disabled flags. It should return a successful result with reason=DISABLED and no value/variant, matching the updated flagd core behavior.

Changes

• FlagdCore: when a flag has state: DISABLED, return reason=DISABLED with value and variant omitted
• Targeting rules should not be evaluated for disabled flags
• Verify behavior against the flagd-testbed scenarios once feat: add disabled flag evaluation test scenarios flagd-testbed#375 lands

[jonathannorris]

Testbed scenarios are up in open-feature/flagd-testbed#376 (pending merge). Once that lands, bump the flagd-testbed submodule and implement the two @in-process scenarios from gherkin/disabled.feature:

1. Evaluating a disabled flag returns reason DISABLED and code default — single-flag eval for boolean, string, integer, float; all reuse existing step definitions
2. Flag disabled in one flag set, enabled in another — uses the selector option to load rawflags/disabled-flags.json (flag DISABLED) vs rawflags/selector-flags.json (flag ENABLED); reuses existing selector steps

The @rpc bulk eval scenario doesn't apply to in-process providers.