Add Feature Engineering reference files to databricks-ml-training-serving

[Hehehe421]

Summary

Extends the databricks-ml-training-serving skill with two new reference files covering Feature Engineering in Unity Catalog.

• 3-feature-store.md (287 lines) — Standard API (GA). Covers FeatureEngineeringClient, create_table/write_table, FeatureLookup with point-in-time, fe.log_model with lineage, score_batch, and online store publishing via Lakebase.
• 4-declarative-features.md (296 lines) — Declarative API (Public Preview, >=0.15.0). Covers DeltaTableSource, create_feature with SlidingWindow/TumblingWindow/RollingWindow, materialize_features, and Feature Serving Endpoints. Streaming Declarative Features (Private Preview) referenced as a callout only — not covered in code.
• SKILL.md — description updated with 14 feature store trigger terms so the skill activates on Feature Engineering prompts (feature store, FeatureLookup, score_batch, publish_table, create_feature, etc.).

Why

The existing skill covered classical ML and MLflow training but did not address Feature Engineering — a core ML pattern for Databricks customers needing training-serving consistency, feature reuse, point-in-time correctness, and real-time lookups.

The existing SKILL.md also referenced `mlflow.pyfunc.spark_udf` as the batch scoring path. For feature-store-backed models this is the wrong API — `fe.score_batch` is required to auto-join feature tables via lineage. The new files document the correct path; the trigger terms in SKILL.md description ensure the agent uses the right one.

Source material

• dbdemos Feature Store notebooks (notebooks 01, 02, 04)
• Declarative Feature Engineering docs
• Feature Engineering in Unity Catalog docs

Evaluation

Evaluated with SkillForge L1 (unit) + L3 (static document quality):

• Composite score: 0.96
• L1: 1.00 (50/50 link and syntax checks pass)
• L3: 0.92 (10-dimension document quality rubric)
• A/B comparison vs baseline: B (with feature store files) wins at 0.95 confidence, no regression flags

Test plan

• Live tested against summer_catalog.dbdemos_fs_travel in a vm-summer workspace (T1-T4 manual prompts via Databricks Assistant)
• All code blocks validated as syntactically correct by SkillForge L1
• All cross-skill references resolve correctly (49/49 link checks pass)
• SKILL.md description change verified to be the only modification to the existing file

[Hehehe421]

Migrating contribution to databricks-eng/universe per the latest agent-skills pipeline guidance. Branch: sixuan-he_data/feature/feature-store-skill on databricks-eng/universe-dev. PR will open against universe:master once Tian's declarative-features.md lands.