fix: XGBoostEstimator honors random_seed config

[immu4989]

Why are these changes needed?

Seed random_state on XGBoostEstimator.__init__ so its native-API xgb.train(self.params, ...) path produces deterministic results across runs and honors the FLAML-internal random_seed config. Uses the same defensive pattern shipped in #1541, #1546, #1547, and #1549 — pop the random_seed key from self.params, and only set random_state when the caller has not already provided one.

XGBoostEstimator is the non-sklearn-API XGBoost class — not bound to a user-facing string in generic_task.py, but reachable via:

• Direct instantiation (e.g. flaml/automl/automl.py uses it internally for the best-config retrain path)
• User subclassing (FLAML's own test_regression.py defines MyXGB1 / MyXGB2 subclasses that exercise this exact pattern)

This is the last XGBoost item on tracking issue #1540; the sklearn-API variants XGBoostSklearnEstimator and XGBoostLimitDepthEstimator were closed by #1549.

Related issue number

Tracking issue: #1540
Pattern reference: #1549 (XGBoostSklearnEstimator), #1547 (RandomForest), #1541 (SGD), #1546 (LRL1)

Checks

• I've used pre-commit to lint the changes in this PR (note the same in integrated in our CI checks).
• I've included any doc changes needed for https://microsoft.github.io/FLAML/. See https://microsoft.github.io/FLAML/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.