fix: use custom provider context for aux compression

[MatthewPJ]

Summary

• persist resolved custom_providers on AIAgent
• pass custom_providers into auxiliary compression context-length detection
• add regression coverage for named custom provider per-model context_length

Test Plan

• /Users/matthew/.hermes/hermes-agent/venv/bin/python -m pytest tests/run_agent/test_compression_feasibility.py tests/agent/test_model_metadata_local_ctx.py -q -o 'addopts='

[alt-glitch]

Likely duplicate of #19564 (and #13540, #18877) — same fix: pass custom_providers into aux compression context-length detection in run_agent.py. See also tracking issue #12977.

[alt-glitch]

Likely duplicate of #19564 (and #13540, #18877) — same fix: pass custom_providers into aux compression context-length detection in run_agent.py. See also tracking issue #12977.

[teknium1]

Thanks for the fix. This is already implemented on current main, so I'm closing this as implemented there.

Automated hermes-sweeper review evidence:

• Current main stores resolved compatible custom providers on the agent for compression feasibility reuse: agent/agent_init.py:1395-1407.
• Current main passes those providers into auxiliary compression context-length detection: agent/conversation_compression.py:136-145 calls get_model_context_length(..., custom_providers=agent._custom_providers).
• get_model_context_length consumes that value before probing and honors per-model custom provider context_length overrides: agent/model_metadata.py:1530-1578.
• The implementing commit is 7becb19ea00c13bdff6f78b71aa3ddfb0bdb5378 (fix(auxiliary): forward custom_providers to compression model context-length detection), which is contained in release tag v2026.5.16.

The prior duplicate comments from @alt-glitch match this code path.