feat: add framework-neutral core toolkit surface

[byjoh]

Summary

• add a framework-neutral core surface for listing Extend tool specs and executing raw tools
• move framework and server dependencies behind optional extras so base installs stay lightweight
• expand endpoint tool coverage for transaction counts, category labels, organizations, users, and expense policies
• document core usage plus adapter extras for custom agent runtimes and service integrations

Details

The new core API exposes structured tool metadata, including stable refs, input schemas, required scopes, action metadata, and read/write classification. It also provides raw structured execution through existing shared functions without requiring callers to use a specific AI framework adapter.

Framework-specific packages are now optional extras for LangChain, MCP, OpenAI Agents, and CrewAI. Top-level imports lazy-load optional adapters so core usage does not require those dependencies.

The catalog now includes additional endpoint families used by service integrations: transaction counting, expense category label detail, organization and user metadata, and organization expense policy raw text. For endpoints not modeled by the pinned SDK client, the toolkit uses a small raw request helper against the SDK API client while preserving the same tool metadata and execution path.

Validation

• python -m pytest extend_ai_toolkit/tests -q
• python -m ruff check --select F extend_ai_toolkit/core.py extend_ai_toolkit/shared/enums.py extend_ai_toolkit/shared/configuration.py extend_ai_toolkit/shared/schemas.py extend_ai_toolkit/shared/prompts.py extend_ai_toolkit/shared/functions.py extend_ai_toolkit/shared/tools.py extend_ai_toolkit/tests/test_endpoint_tool_coverage.py
• python -m build
• git diff --check