fix: preserve config_context_length when switching models

[rarf]

Problem

When switching models at runtime via /model <new-model>, the custom config_context_length override from config.yaml was not being passed to the new ContextCompressor instance. This caused the user-specified context length to be lost after a model switch.

Fix

• Store _config_context_length on AIAgent instance during __init__
• Pass _config_context_length when creating new ContextCompressor in switch_model()
• Add test to verify config_context_length is preserved across model switches

Testing

• Added tests/run_agent/test_switch_model_context.py with 2 test cases:
  • test_switch_model_preserves_config_context_length - verifies config override is preserved
  • test_switch_model_without_config_context_length - verifies behavior when no override is set

Fixes: quando estamos alterando o modelo não está alterando o tamanho do contexto

[trevorgordon981]

In run_agent.py (lines 1145-1386) and hermes_cli/model_switch.py (lines 790-811), changes are made to preserve config_context_length across model switches and provider listings.

Observation: The switch_model method (lines 1383-1386) now explicitly passes config_context_length to the context compressor. However, in list_authenticated_providers (lines 800-811), the logic checks overlay.extra_env_vars and api_key_env_vars from PROVIDER_REGISTRY to determine credential presence.

Question: In list_authenticated_providers, if a provider (like opencode-go) relies on api_key_env_vars defined in PROVIDER_REGISTRY rather than extra_env_vars in HERMES_OVERLAYS, is the current logic robust enough to handle all auth types correctly? Specifically, if api_key_env_vars are updated or added to PROVIDER_REGISTRY after this code is written, will this function automatically adapt, or does it rely on hardcoded assumptions about which env vars are checked?

[teknium1]

Merged via PR #7187. Both commits cherry-picked with your authorship preserved. Thanks for both fixes!