Add OpenFeature provider

[1fanwang]

Why

Airflow has no vendor-neutral way to evaluate feature flags. Teams that want to canary or
progressively roll out a change (a new operator behavior, a migration, an executor option) either
hard-code a specific vendor SDK inside a PythonOperator or write a bespoke hook per backend.
OpenFeature is the CNCF standard for exactly this: one evaluation API with
a pluggable provider per backend (flagd, GrowthBook, Unleash, an in-house system).

What

A new apache-airflow-providers-openfeature provider:

• OpenFeatureHook — registers an OpenFeature provider from a connection and evaluates
  boolean/string flags with a targeting context.
• FeatureFlagSensor — waits until a flag is enabled for a targeting entity, e.g. hold a task until a
  rollout reaches its dag_id.
• FractionalProvider — a dependency-free, in-process provider for deterministic percentage rollouts,
  so a canary works with no external flag daemon (mirrors flagd's fractional op).

The hook also backs a cluster-policy pattern (documented in the provider index): consult a flag from
task_policy in airflow_local_settings.py to flip a cohort's pool/queue and ramp a platform
change like a canary, with no core change. That progressive-delivery use case is the point; the
sensor and hook are the building blocks.

Scope is intentionally small to open the discussion. Out of scope here: vendor-specific provider
adapters (those belong to each vendor or the user's own code), named-client multi-provider setups, and
any core change.

Tests

providers/openfeature/tests/unit — 11 unit tests: the fractional provider (determinism, ramp
distribution, weighted variants), the hook (global-provider fallback, provider registration from a
connection, register-once), and the sensor (enabled, disabled, expected). Green on Airflow 3.2.2 with
openfeature-sdk 0.10.0.

Risk

Additive and opt-in: a new provider, nothing existing changes. openfeature-sdk is the only new
dependency and is pulled only when the provider is installed. FractionalProvider is a convenience for
testing and simple rollouts; production points OpenFeature at a real backend.

---

Was generative AI tooling used to co-author this PR?

• Yes (please specify the tool below)

Generated-by: GitHub Copilot CLI following the guidelines