Add image_generation server-side tool

[yanxi0830]

Summary

Adds the server-side image_generation tool to the SDK, mirroring the tool now supported by the xAI API (following the web_search / x_search / code_execution agentic-tool pattern):

• xai_sdk.tools.image_generation(): new helper that returns a chat_pb2.Tool for image generation and editing in agentic requests. Takes an optional action parameter — "auto" (default; generate + edit), "generate" (text-to-image only), or "edit" (image editing only). Omitting action leaves the field unset so the server applies its default.
• Regenerated protos (v5 & v6):
  • chat.proto: ImageGeneration message (optional string action = 1), image_generation = 10 in the Tool oneof, and TOOL_CALL_TYPE_IMAGE_GENERATION_TOOL = 10.
  • usage.proto: SERVER_SIDE_TOOL_IMAGE_GENERATION = 11 and num_image_generations / num_image_edits fields on SamplingUsage for per-call generation/edit billing visibility.
• get_tool_call_type() docstring now lists image_generation_tool as a possible return value.
• Examples: examples/sync/image_generation_tool.py and examples/aio/image_generation_tool.py showing generation, decoding the ROLE_TOOL result envelope ({"__type": "image_generation_result", "result": "data:<mime>;base64,..."}) to save the image to disk, streaming, and multi-turn editing via previous_response_id.

Usage:

from xai_sdk.tools import image_generation

chat = client.chat.create("grok-4", tools=[image_generation()])
# or restrict to text-to-image only:
chat = client.chat.create("grok-4", tools=[image_generation(action="generate")])

Test plan

• uv run ruff format --check passes
• uv run ruff check passes
• uv run pyright passes (0 errors)
• uv run pytest -n auto — all 770 tests pass
• New unit tests: image_generation() default and per-action proto construction, TOOL_CALL_TYPE_IMAGE_GENERATION_TOOL mapping via get_tool_call_type, SERVER_SIDE_TOOL_IMAGE_GENERATION enum value, SamplingUsage image fields, and sync/async chat.create(tools=[...]) round-trips
• Verified the regenerated v5 protos import and work under protobuf==5.29.4, and v6 under protobuf==6.31.1
• Live end-to-end verification against the production API: tool call surfaces as imagine_text_to_image / imagine_image_to_image with TOOL_CALL_TYPE_IMAGE_GENERATION_TOOL, the ROLE_TOOL envelope decodes to a valid image, usage reports SERVER_SIDE_TOOL_IMAGE_GENERATION with num_image_generations / num_image_edits populated, and both example scripts (sync generate, multi-turn edit, async streaming) run successfully