Bug Report: Groq models trigger false "Context overflow" error

Summary

OpenClaw incorrectly calculates context limits for Groq models, causing immediate "Context overflow" errors even with minimal workspace context (~12KB) and simple messages. The error is mislabeled as a "billing" error, which then disables the auth profile.

Environment

• OpenClaw version: 2026.2.1 (ed4529e)
• OS: macOS Darwin 22.6.0
• Node: v22.22.0
• Groq models tested: groq/llama-3.3-70b-versatile, groq/llama-3.1-8b-instant

Steps to Reproduce

1. Configure Groq as primary model:

   openclaw models set groq/llama-3.3-70b-versatile

2. Add Groq API key to auth-profiles.json

3. Send any WhatsApp message to trigger agent response

4. Observe immediate "Context overflow" error

Expected Behavior

• Groq llama-3.3-70b-versatile has 131,072 token context window (verified via Groq API)
• Groq llama-3.1-8b-instant has 131,072 token context window
• A ~12KB workspace + simple message should easily fit

Actual Behavior

FailoverError: Context overflow: prompt too large for the model.
Try again with less input or a larger-context model.

Error occurs within 248ms of request start, before any API call is made.

The error is then mislabeled as (billing) which disables the auth profile:

"usageStats": {
  "groq:default": {
    "errorCount": 1,
    "failureCounts": { "billing": 1 },
    "disabledReason": "billing"
  }
}

Verification: Direct Groq API Works

curl -s https://api.groq.com/openai/v1/chat/completions \
  -H "Authorization: Bearer $GROQ_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "llama-3.3-70b-versatile", "messages": [{"role": "user", "content": "Say hello"}]}'

Response: Works correctly, returns "Hello! How can I assist you today?"

Attempted Workarounds (All Failed)

1. Trimmed workspace files from 32KB to 12KB - still fails
2. Cleared all session history - still fails
3. Fresh gateway restart - still fails
4. Cleared auth profile usageStats - profile gets disabled again on first request

Root Cause Hypothesis

OpenClaw appears to have a hardcoded or incorrectly configured context limit for Groq models that doesn't match their actual 131K context windows. The context calculation happens client-side before the API request.

Logs

[diagnostic] lane task error: lane=main durationMs=248
error="FailoverError: Context overflow: prompt too large for the model."

[diagnostic] lane task error: lane=session:agent:main:whatsapp:dm:+12177374277
durationMs=295 error="FailoverError: Context overflow..."

Embedded agent failed before reply: All models failed (2):
groq/llama-3.3-70b-versatile: Context overflow... (billing) |
groq/llama-3.1-8b-instant: No available auth profile for groq (all in cooldown)

Impact

• Groq integration is completely unusable
• Users cannot use free Groq API tier with OpenClaw
• Auth profiles get permanently disabled until manually cleared

Suggested Fix

Update Groq model context limits in OpenClaw's model registry to match actual values:

• llama-3.3-70b-versatile: 131,072 tokens
• llama-3.1-8b-instant: 131,072 tokens
• Other Groq models: Check https://api.groq.com/openai/v1/models for context_window field