enhancement: evaluate AI context provider upload adapters

[codeforester]

Summary

Evaluate whether Base can safely support provider upload adapters for AI context
bundles after basectl export-context exists.

The goal is to verify the real supported API surfaces for Claude, ChatGPT, and
future providers such as Gemini before Base commits to provider-specific upload
behavior.

Current Understanding

• Anthropic's Files API supports uploading files for API calls, but does not
  clearly provide a normal write path into Claude.ai Projects.
• Claude.ai project content appears in Enterprise Compliance API documentation
  as retrievable/deletable content, not as a general upload target.
• OpenAI's Files API supports API features such as assistants, fine-tuning,
  batch, vision, and user data.
• OpenAI's Projects API manages API/organization projects, not ChatGPT UI
  project knowledge uploads.

This issue should refresh those assumptions against current official docs when
implementation time arrives.

Acceptance Criteria

• Document official upload/write support for Claude project knowledge, ChatGPT
  project knowledge, and Gemini/project-equivalent context.
• Identify which providers are safe to support in Base, if any.
• If no provider has a supported write API, document the manual workflow that
  basectl export-context should support.
• If a provider does have a supported write API, create follow-up implementation
  issues with exact API endpoints, authentication model, config shape, and
  security constraints.
• Confirm API keys/tokens would only be read from environment variables or
  platform-native auth, never stored in repo files.

Out of Scope

• Implementing provider upload adapters.
• Browser automation for manual uploads.
• Storing provider credentials in ~/.base.d/config.yaml.