feat: add AgentSkill type and skills support to generate methods

[devin-ai-integration]

feat: add AgentSkill type and skills support to generate methods

Summary

Adds AgentSkill model and skills support across prediction methods, allowing callers to use skill_name + version as an alternative to specifying a domain. This aligns the Python SDK with the backend changes in vlm-lab#1621.

Changes:

• New AgentSkill Pydantic model with skill_id, skill_name, version, type fields and a validator requiring at least one identifier
• Added skills field to GenerationConfig and AgentExecutionOrCreationConfig
• Made domain optional in ImagePredictions.generate() and FilePredictions.generate() — now either domain or config.skills must be provided
• domain key is only sent in the request payload when explicitly provided
• autocast is skipped when no domain is present (skills-based flow)

Review & Testing Checklist for Human

• Forward reference ordering: AgentExecutionOrCreationConfig (line ~230) uses List[AgentSkill] in its type annotation, but AgentSkill is defined later (~line 315). Verify this doesn't cause a NameError at import time — this depends on whether from __future__ import annotations is present at the top of types.py or if Pydantic defers annotation resolution. This is the highest-risk item.
• Removed validation block in ImagePredictions.generate(): A large block of manual image/URL validation was replaced by a call to self._handle_images_or_urls(). Verify that _handle_images_or_urls covers all the same edge cases (empty lists, mixed types, non-http URLs, etc.)
• Serialization of skills in model_dump(): GenerationConfig.model_dump() needs to correctly serialize nested AgentSkill objects into the JSON payload the backend expects (skill_id, skill_name, version). Test with an actual API call or at minimum a unit test.
• End-to-end test: Make a real API call using config=GenerationConfig(skills=[AgentSkill(skill_name="...", version="latest")]) without a domain to verify the backend accepts it and returns the expected response.

Notes

• CI checks passing is necessary but insufficient — the compilation test only validates syntax, not runtime behavior
• This PR is part of a coordinated update across Python SDK, Node SDK, and docs
• Link to Devin run: https://app.devin.ai/sessions/6d9ce36452de49339048a051a670858e
• Requested by: dinesh@vlm.run

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring