Introduce BaseAIHook to common-ai provider

[gopidesupavan]

Introduces BaseAIHook, an abstract hook that defines the contract every
agent-framework backend must implement. AgentOperator (and the @task.agent
decorator) now selects the backend at runtime from the Airflow connection's
conn_type — no new operator class is needed when adding another framework.

What changed

New: BaseAIHook contract (hooks/base_ai.py)

• BaseAIHook(BaseHook) — abstract base with three abstract methods:
  • get_conn() — return the backend model/client
  • create_agent(output_type, instructions, **kwargs) — build an agent
  • run_agent(agent, *, prompt, usage_limits, message_history) → AgentRunResult
• AgentRunResult dataclass — backend-neutral result:
  output, message_history, model_name, usage, tool_names
• AgentUsage dataclass — normalized token/request counters:
  requests, tool_calls, input_tokens, output_tokens, total_tokens
• Class-level capability flags: supports_toolsets, supports_durable,
  supports_usage_limits
• BaseAIHook.get_agent_hook(conn_id) — resolves the registered hook from
  conn_type and raises TypeError if it is not a BaseAIHook

Refactor: PydanticAIHook(BaseAIHook)

PydanticAIHook (and its subclasses PydanticAIAzureHook,
PydanticAIBedrockHook, PydanticAIVertexHook) now subclass BaseAIHook.
run_agent() wraps the pydantic-ai RunResult into AgentRunResult, so all
callers receive a backend-neutral object.

AgentOperator dispatch via get_agent_hook()

AgentOperator.execute() calls BaseAIHook.get_agent_hook(conn_id) instead of
importing PydanticAIHook directly. This is the main extensibility seam:
a future hook registered under a new conn_type is picked up with no operator
changes.

LLMOperator / LLMFileAnalysisOperator aligned

Both operators previously called agent.run_sync() directly, bypassing the hook
abstraction and receiving a pydantic-ai RunResult rather than an
AgentRunResult. They now call self.llm_hook.run_agent(agent, prompt=…),
which means:

• log_run_summary() receives a consistent AgentRunResult from all operators
• The operators are backend-agnostic alongside AgentOperator

logging.py backend-agnostic

log_run_summary() now reads result.model_name, result.usage.*, and
result.tool_names from AgentRunResult directly — no more pydantic-ai
result.response.model_name or result.usage() callable.

Tests

• New tests/unit/common/ai/hooks/test_base_ai.py covering the contract,
  get_agent_hook() dispatch, and TypeError on non-agent hook
• Updated test_pydantic_ai.py, test_agent.py, test_llm.py,
  test_llm_file_analysis.py to mock hook.run_agent() returning
  AgentRunResult instead of a raw pydantic-ai RunResult

---

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

• Yes (please specify the tool below)

---

• Read the Pull Request Guidelines for more information. Note: commit author/co-author name and email in commits become permanently public when merged.
• For fundamental code changes, an Airflow Improvement Proposal (AIP) is needed.
• When adding dependency, check compliance with the ASF 3rd Party License Policy.
• For significant user-facing changes create newsfragment: {pr_number}.significant.rst, in airflow-core/newsfragments. You can add this file in a follow-up commit after the PR is created so you know the PR number.