fix: return conservative token limits for unspecified models

[LLuke]

to avoid crashes

Summary

• When working with a model not specially handled and coded in rust/crates/api/src/providers/mod.rs, the claw doesn't have an idea about the context window size. It will keep on growing without automatic compact, eventually lead of crash of the claw process when an error is thrown in the main.rs. The proper way fix is configure and don't hardcode any. However, we still need a reasonable default du jour to reduce chance of crashes. Thus it is picked to be max_output_tokens: 16_384, context_window_tokens: 131_072 as the minimum of each among the current specified models. While this doesn't prevent crash while working with much less capable models and may under-utilize much more capable models, it is a line in the sand currently needed before claw-code further optimize.

Anti-slop triage

• Classification: actionable-fix
• Evidence: error: Context window blocked #3062
• Non-destructive review result: merge candidate

Verification

• Targeted tests/docs checks ran, or the gap is explicitly recorded.
• git diff --check passes.
• No live secrets, tokens, private logs, or unrelated generated churn are included.

Resolution gate

• If this PR resolves an issue, the issue number and fix evidence are linked.
• If this PR should not merge, the rejection/defer rationale is evidence-backed and does not rely on vibes.
• I did not merge/close remote PRs or issues from an automation lane without owner approval.

[LLuke]

After the fix, testing clawd with MiniMax M2.5 with up to 192K token context window is stable with auto-compacted seen:

...
[auto-compacted: removed 61 messages]
...
[auto-compacted: removed 8 messages]

The 16K/132K were much lower than the model's full capacity but as a safe default the clawd code no longer hangs or crash, -- it is rather snappy.