Add list_all_* helpers that drain pagination on the client

[adityasingh2400]

Closes #2556.

Summary

• Add list_all_tools / list_all_prompts / list_all_resources / list_all_resource_templates on Client. Each walks next_cursor until exhausted and returns the combined list.
• Add iter_all_tools / iter_all_prompts / iter_all_resources / iter_all_resource_templates async iterators, for streaming consumers that don't want to materialize every page in memory.
• Update the docstrings on the single-page list_* methods to say so plainly and point at the new drains.
• Switch ClientSessionGroup._aggregate_components to drain pagination so multi-server aggregators don't silently drop pages past the first (this was specifically called out in the issue as the scenario where high tool counts are most likely).

Existing tests that mocked the session's list_tools / list_resources / list_prompts now set next_cursor=None on the mock return values so the new drain loop terminates after a single page.

Why

The single-page list_* methods are easy to misuse, which is the whole point of the issue:

Connected Claude Code to an AWS Bedrock AgentCore Gateway with 31 tools registered across 7 targets. Only 14 showed up. The gateway returns 14 on page 0 with a nextCursor, then 17 on page 1. Claude Code calls tools/list once and drops the cursor.

The existing primitives are kept; this just adds a drain helper next to them so downstream client code can get the "give me everything" behavior in one call instead of writing the loop themselves.

Test plan

• uv run pytest tests/client/test_list_all_pagination.py tests/client/test_session_group.py tests/client/test_list_methods_cursor.py (25 passed)
• uv run pytest tests/client/ (208 passed, 1 xfailed)
• uv run pytest tests/ (1180 passed, 98 skipped, 1 xfailed)
• uv run ruff check . && uv run ruff format --check . (clean)
• uv run pyright src/mcp/ tests/client/test_list_all_pagination.py tests/client/test_session_group.py (0 errors)
• 100% coverage on src/mcp/client/client.py and src/mcp/client/session_group.py via tests/client/

New tests cover:

• Multi-page drain across 2 and 3 pages for tools, prompts, resources, and resource templates.
• The cursor sent on subsequent requests matches the next_cursor returned by the previous page.
• Single-page and empty-server cases.
• iter_all_tools yields one tool at a time and pages lazily.
• ClientSessionGroup aggregates across paginated list_tools calls and supplies the cursor on the follow-up request.

Notes

I went with Client helpers only and left ClientSession as the single-page primitive (matches what the issue suggested as the conservative call). Happy to also add session-level helpers if you'd prefer that.