fix: [Price] deepseek-v4-pro cost doesn't match DeepSeek's official pricing page

[github-actions]

Summary

Corrects deepseek-v4-pro pricing on the first-party DeepSeek provider. Catalog values were stale/mis-converted and did not match DeepSeek’s published off-peak USD rates. Aligns with the same off-peak convention already used for deepseek-flash / deepseek-v4-flash.

Changes

File: providers/deepseek/models/deepseek-v4-pro.toml

Field	Old	New
cost.input	0.435	0.66
cost.output	0.87	1.98
cost.reasoning	0.87	1.98
cost.cache_read	0.003625	0.022

Also refreshed the leading source comment to note the off-peak convention and access date 2026-09-19.

Evidence

• DeepSeek Models & Pricing (EN) — for deepseek-v4-pro (DeepSeek-V4-Pro-0813), off-peak USD per 1M tokens:
  • cache hit (input): $0.022
  • cache miss (input): $0.66
  • output: $1.98
  • Peak is 2× off-peak; models.dev records off-peak (same as flash entries).
• DeepSeek 模型 & 价格 (ZH) — same model off-peak CNY: ¥0.15 / ¥4.5 / ¥13.5 (consistent with EN USD list).
• Peer convention: providers/deepseek/models/deepseek-flash.toml already uses official off-peak USD (0.15 / 0.6 / 0.003) with an explicit off-peak comment.

Validation

• Compared current TOML costs against live DeepSeek EN and ZH pricing pages (fetched 2026-09-19).
• Confirmed EN page publishes USD directly (no FX conversion needed).
• Confirmed flash provider entries already follow the off-peak USD convention.
• Did not run bun validate (shell disabled in this workflow).

Review notes

• Peak/off-peak split still cannot be expressed in schema ([Feat] Support time-based (peak/off-peak) pricing for DeepSeek V4 Pro and Flash #5218); catalog keeps off-peak only, matching flash.
• Scope limited to first-party providers/deepseek/models/deepseek-v4-pro.toml as reported in the issue; other hosts’ prices left unchanged.

Closes #7494

Automated by the issue fixer: https://github.com/anomalyco/models.dev/actions/runs/35435320735