feat(types): export EffortLevel as a named type alias

[bugbubug]

Summary

Promote the inlined Literal["low", "medium", "high", "xhigh", "max"] used by the effort field into a named alias EffortLevel, mirroring the existing pattern used for PermissionMode, SdkBeta, SettingSource, etc.

Closes #938.

Motivation

The effort literal is currently inlined at two sites in types.py:

• AgentDefinition.effort (line 99)
• ClaudeAgentOptions thinking effort (line 1867)

Every other small literal of the same shape in this file is already exported as a named alias. Inlining effort makes the SDK inconsistent with itself and forces downstream wrappers (e.g. internal agent frameworks, third-party libraries) to either duplicate the literal or maintain a private alias that drifts on every SDK release.

The literal is in fact moving: "xhigh" was added between 0.1.71 and 0.1.77 (see #834). A named alias makes any future expansion propagate automatically to all downstream consumers via from claude_agent_sdk import EffortLevel.

Changes

• src/claude_agent_sdk/types.py
  • Add EffortLevel = Literal["low", "medium", "high", "xhigh", "max"] near the other top-level aliases.
  • Replace the inlined literal at the two existing use sites with EffortLevel.
• src/claude_agent_sdk/__init__.py
  • Re-export EffortLevel from the package alongside PermissionMode, SettingSource, SdkBeta, etc.

The alias follows the same Name = Literal[...] style as PermissionMode / SdkBeta / SettingSource rather than the explicit : TypeAlias annotation, to stay consistent with the surrounding file.

Verification

Ran the full local quality gate per CLAUDE.md:

python -m ruff check src/ tests/        # All checks passed!
python -m ruff format --check src/ tests/   # 54 files already formatted
python -m mypy src/                     # Success: no issues found in 24 source files
python -m pytest tests/                 # 743 passed, 5 skipped, 2 failed

The 2 failing tests (test_transcript_mirror.py::TestBuildMirrorBatcherFlushMode::test_eager_mode_flushes_per_frame and TestReceiveLoopFramePeeling::test_eager_flush_mode_appends_per_frame_before_result) also fail on main at the current HEAD without these changes, so they are pre-existing and unrelated.

Also smoke-tested the new public export:

>>> from claude_agent_sdk import EffortLevel
>>> EffortLevel
typing.Literal['low', 'medium', 'high', 'xhigh', 'max']

Notes

No runtime behavior changes — this is a pure type-only refactor. No tests were added since the change is a structural rename of an existing Literal and is covered by mypy strict on src/.