feat(ai): Architecture for High-Density Agent Context (Reducing Token Tax by 60%)

[merchantmoh-debug]

Is your feature request related to a problem? Please describe.

The Problem: The "Hidden Tax" on AI Orchestration
As Go adoption grows for Agentic AI workflows (Gemini/GPT-4o), passing raw go-github structs to LLMs incurs a massive "Noise Tax."
Currently, a standard Issue or PullRequest struct contains HATEOAS links, internal NodeIDs, and redundant pointers. This wastes ~60% of the LLM context window.

For enterprise users building RAG/Agents on Google Cloud, this results in:

1. High Latency: Slower ingestion times.
2. High Cost: Wasted spend on input tokens for metadata the LLM doesn't need.
3. Hallucinations: Lower Signal-to-Noise ratio distracts the model.

Describe the solution you'd like

The Solution: ToAgentContext() (The Silicon Diet)
I propose adding a high-density serialization method ToAgentContext() for core entities (Issue, PullRequest, Repository, Comment).

Proposed Behavior:

• Transform "Fat Structs" into map[string]any optimized for RAG.
• Strip all HATEOAS, URLs, and non-narrative metadata.
• Enforce strict token limits on body content to prevent context overflow.

Strategic Value:
This enables Go to compete directly with Python frameworks for high-efficiency AI agents by treating Context as a scarce resource.

Describe alternatives you've considered

Alternative: Users manually marshal/unmarshal structs and scrub data.
Downside: Error-prone, boilerplate-heavy, and prevents standardization across the ecosystem.

Additional context

Benchmarks (Verified in PR #3967):

• Standard Payload: ~2,000 bytes
• Agent Context: ~790 bytes
• DENSITY IMPROVEMENT: ~60% 📉

Implementation:
A full implementation is already submitted and passing all tests in PR #3967.