Python: Defer experimental AgentFactory creation in WorkflowFactory - #10114

Motivation & Context

Constructing WorkflowFactory() eagerly creates an experimental AgentFactory, so even a workflow containing only SendActivity emits a DECLARATIVE_AGENTS warning. Applications using the stable workflow APIs should enter the declarative-agent loading path only when their workflow requires it.

Description & Review Guide

• What are the major changes? Lazily initialize and cache the default agent factory when processing an agent definition. Document the warning boundary and the pre-registered agent option. Add nine regression cases covering ordinary workflows with and without an environment file, both registration methods, inline/file-based agent definitions, factory reuse, and an explicitly supplied agent factory.
• What is the impact of these changes? Action-only workflows and workflows using pre-registered agents can be constructed and run with experimental warnings treated as errors. YAML agent loading emits the experimental warning on first use. When no custom agent factory is supplied, WorkflowFactory loads env_file during initialization independently of agent creation, preserving =Env.NAME values for workflows that enable process-environment fallback with restrict_env_to_configuration=False. Subsequent agent definitions reuse the default agent factory.
• What do you want reviewers to focus on? The lazy initialization boundary, registered-agent precedence, preserved environment-file loading, and the isolated once-per-feature warning state in the regression tests.

Related Issue

Fixes #10092.

The issue has the ready-for-implementation label. No competing open PR referencing #10092 was found before submission.

AI Assistance

• No material AI assistance was used.
• This is an AI-assisted contribution.

Codex assisted with issue analysis, implementation, regression tests, documentation, and this PR description. I have reviewed all submitted content and take resonsibility for it.

Contribution Checklist

• The code builds clean without any errors or warnings
• All unit tests pass, and I have added new tests where possible
  • Declarative package on Python 3.13 with the Power Fx runtime available — 1677 passed, 4 skipped, 94% coverage. The suite reports eight existing MCP deprecation warnings.
  • Ruff lint/format, strict source Pyright, and all five configured type checkers for the changed test files passed.
  • The new regression cases reproduced five failures before the implementation change.
• The PR follows the Contribution Guidelines
• This PR links to an agreed issue with no competing open PR, or the Related Issue section documents a trivial-change or repository-automation exception.
• This is not a breaking change. If it is a breaking change, add the breaking change label (or add "[BREAKING]" to the title prefix, before or after any language prefix) — a workflow keeps the label and title prefix in sync automatically.