feat(custom-providers): add context_cache setting to disable context length caching

[gkraker04]

What does this PR do?

Adds a context_cache boolean field to custom_providers config that controls whether detected context lengths are persisted to ~/.hermes/context_length_cache.yaml.

Problem: When using custom providers (local llama.cpp, Ollama, vLLM servers), Hermes caches the detected context length to avoid repeated API calls on startup. While efficient, this causes issues when:

• Server configuration changes (e.g., Ollama num_ctx parameter)
• Models are swapped with different context limits
• Users are debugging context detection issues
• Containerized environments where context size varies between deployments

Solution: Add context_cache field (default true for backward compatibility) that can be set at:

1. Provider level - affects all models on that endpoint
2. Model level - granular control per model, overrides provider default

When context_cache: false, the persistent cache lookup is skipped and Hermes performs fresh detection on every startup via endpoint queries, local server APIs, or models.dev registry.

Related Issue

No existing issue - this is a new feature request from community feedback.

Type of Change

• ✨ New feature (non-breaking change that adds functionality)

Changes Made

• agent/model_metadata.py (+13, -3): Added context_cache parameter to get_model_context_length(), updated cache lookup logic
• agent/context_compressor.py (+2): Added parameter and passed through to get_model_context_length()
• run_agent.py (+16, -2): Read context_cache from custom_providers config (provider and model level), passed to ContextCompressor
• tests/agent/test_context_cache.py (new, 184 lines): 7 comprehensive tests covering all scenarios

How to Test

# Run new tests
python -m pytest tests/agent/test_context_cache.py -v

# Verify no regression in existing tests
python -m pytest tests/agent/test_model_metadata.py -v

All 75 existing tests pass + 7 new tests added.

Usage example:

custom_providers:
  - name: "local-dev"
    base_url: "http://localhost:8080/v1"
    context_cache: false  # Always re-detect all models
    models:
      qwen3.5:27b:
        context_cache: true  # Per-model override

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (feat(scope):, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this feature (no unrelated commits)
• I've run `pytest tests/