[linqiang-max]

Fixes #1013

xAI lists all three Grok 4.20 models at $1.25 input / $0.20 cached / $2.50 output per 1M tokens for prompts under 200K tokens: https://docs.x.ai/developers/pricing (checked 2026-11-09). The file had $2 / $6.

Changed request_token 0.0002 → 0.000125 and response_token 0.0006 → 0.00025 for:

• grok-4.20-0309-reasoning, grok-4.20-0309-non-reasoning, grok-4.20-multi-agent-0309
• their aliases: grok-4.20, grok-4.20-0309, grok-4.20-reasoning, grok-4.20-non-reasoning, grok-4.20-multi-agent, and the -latest variants

The cached rate (0.00002) was already right, and I left additional_units alone.

Not changed:

• the beta / experimental ids and the -gv2 ids. xAI doesn't list them today, so I don't know what they were billed at.
• the long-context tier. At 200K prompt tokens or more, xAI bills the whole request at $2.50 / $0.40 / $5.00. I didn't see a way to express that in this file, so this PR only fixes the standard rate.