test: cover null and empty targeting keys in shared fixtures

[leoromanovsky]

Motivation

The common fixtures' existing empty-targeting-key case uses a static flag with no shards or rules, so it cannot catch an evaluator treating "" as missing during sharding or implicit id resolution. This gap surfaced while updating DataDog/dd-trace-dotnet#8384.

Changes

• Add a sharded flag and cases for an empty targeting key (SPLIT) versus null (ERROR / TARGETING_KEY_MISSING).
• Add an empty-id rule and cases for empty-key fallback (TARGETING_MATCH), null-key fallback (DEFAULT), and explicit id attributes taking precedence.
• Document the six new cases in the fixture index.

Decisions

• Use the existing evaluation-cases/*.json format so current canonical-fixture runners discover the cases.
• Use a full-range shard to require hashing without tying this regression to a particular bucket, and ^$ to match only an empty id.
• Keep the existing static and null-key fixtures. Consumers must preserve null instead of converting it to an empty string; the .NET runner correction is included in fix(openfeature): accept empty string as valid targeting key dd-trace-dotnet#8384.

Validation

• python3 ci/validate-fixtures.py: 313 cases across 42 files passed.
• Loaded the proposed JSON into dd-trace-dotnet at 66c3df873c (fix(openfeature): accept empty string as valid targeting key dd-trace-dotnet#8384): all 571 FeatureFlags tests passed on .NET 10, including the six new cases. Temporary snapshot changes were restored afterward.
• Other SDK behavior remains subject to downstream validation; this local run does not establish cross-SDK compatibility.

Checklist

• JSON files are valid.
• New cases reference flags in ufc-config.json.
• New flags have corresponding evaluation cases.
• Expected results validated against the corrected .NET evaluator.