fix(agent): restore global model.context_length override on /model switch

[ech0hol]

What does this PR do?

When a user sets a global model.context_length in config.yaml and later switches models mid-session via /model, the global override is silently dropped and never takes effect again for the remainder of the session.

Root cause: switch_model() unconditionally clears agent._config_context_length = None at line 1411 to prevent stale per-model overrides from leaking to the new model. But after the swap, the code only re-reads custom_providers (per-model overrides, priority 0b) from config — it never restores the global model.context_length (priority 0).

Fix: After load_config() and get_compatible_custom_providers() in the existing try/except block, read model.context_length from the fresh config and write it back to agent._config_context_length. This is the same value that agent_init.py passes to the agent at startup, so the behaviour after a switch matches a fresh session.

Related Issue

Fixes #40979

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• agent/agent_runtime_helpers.py — in switch_model(), restore the global model.context_length override from load_config() after the per-model custom_providers overrides have been re-read (lines 1523–1534, +12 lines, no deletions)

The new code sits inside the existing try/except Exception block so a malformed config falls through gracefully without blocking the switch.

How to Test

1. Add a global context length override to ~/.hermes/config.yaml:

   model:
     context_length: 64000

2. Start a session with any model (e.g. gpt-4o).
3. Verify the compressor has the override (/debug or check state): context length = 64000.
4. Run /model claude-sonnet-4-20250514 (or any other model).
5. Verify the compressor still reports context length = 64000 — before this fix, it would fall back to the model's default.

Run the relevant tests:

pytest tests/run_agent/test_switch_model_context.py \
       tests/run_agent/test_switch_model_rollback.py \
       tests/run_agent/test_switch_model_fallback_prune.py -v

All 10 tests pass with no regressions.

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix:, feat:, etc.)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix/feature (no unrelated commits)
• I've run pytest tests/run_agent/test_switch_model_*.py and all tests pass
• I've added tests for my changes — existing tests already cover both the "old override must not leak" and "no override = None" paths; the fix is a config-read restoration that is transparent to the test suite
• I've tested on my platform: Windows 11