feat: include selected variant key in flag results

[gagantrivedi]

Changes

Every flag result now carries a variant field, mirroring OpenFeature's ResolutionDetails.variant (always present, nullable):

• the variant's key when a named multivariate variant is selected,
• "control" when an identity falls in the control bucket (the leftover allocation 100 − Σ weights that resolves to the feature's default value) — the control is not a multivariate option, so the engine synthesises this key,
• null otherwise — a standard feature, an unkeyed variant, or evaluation without an identity (nothing is bucketed).

{ "name": "checkout_button", "enabled": true, "value": "blue", "reason": "SPLIT; weight=30", "variant": "v_a" }
{ "name": "checkout_button", "enabled": true, "value": "blue", "reason": "DEFAULT",          "variant": "control" }
{ "name": "basic_flag",      "enabled": false, "value": "",    "reason": "DEFAULT",          "variant": null }

Selection logic, hashing and ordering are unchanged. feat: commit → release-please cuts a minor (10.2.0).

Files

• flag_engine/context/types.py (generated): FeatureValue gains key: NotRequired[str]
• flag_engine/result/types.py (generated): FlagResult.variant: Optional[str] (required, nullable)
• flag_engine/segments/evaluator.py: single variant value threaded through one result construction
• .gitmodules + submodule: corpus bumped to v3.9.0

Test coverage

The behaviour is covered end-to-end by the shared engine-test-data corpus (v3.9.0), so the variant-specific unit test was dropped as redundant:

scenario	variant	corpus
no identity / standard feature	null	✓
unkeyed variant selected	null	✓ (SPLIT cases)
control bucket	"control"	✓
keyed variant selected	the key	✓ (added in v3.9.0)

Landed dependencies

• feat(sdk): make evaluation result variant required and nullable flagsmith#7726 — evaluation result schema variant required + nullable (merged).
• Flagsmith/engine-test-data — variant on every result + the keyed-variant case, released as v3.9.0 (merged); submodule pins the tag.

How did you test this code?

Full suite 382 passed against the released v3.9.0 corpus. mypy (strict), black, flake8, isort clean. Generated types verified with datamodel-code-generator 0.33.0 against the merged schema.

Cross-engine note

flagsmith-rust-flag-engine pins an older corpus tag and is unaffected until it adopts v3.9.0; the keyed-variant case is what will catch key passthrough not being implemented there.

[github-actions]

File	Coverage		Missing
All files	100%	✅

Minimum allowed coverage is 100%

Generated by 🐒 cobertura-action against 871d63c

[codspeed-hq]

Merging this PR will not alter performance

✅ 1 untouched benchmark

---

_{Comparing feat/multivariate-variant-key (871d63c) with main (ef90a90)}