Add MiMo V2.5 Pro (Xiaomi) to the board

[lakshvantb]

Board rows for mimo-v2.5-pro, generated with generate_model_rows.py --append (the PR #6 mode) from the eval-run data on eval/mimo-v2.5-pro-20260721-1654 — this is the live validation of the automated board-PR path (livebench-private #172).

category	score
math	76.0
coding	68.1
data_analysis	48.7
instruction_following	39.7
language	62.9
reasoning	66.0
agentic_coding_v2	41.4
overall	57.5

Cost (Xiaomi first-party pricing 0.435/0.0036/0.87 per M): $0.0166/question, $0.0289/successful task — cheapest model on the board.

Verified: generator self-verify (all columns), check_data.js 43/43 with cost, no orphans, full 1270-question coverage, modelLinks entry (openweight+reasoner).

Merge after the data PR (livebench-private #170); deploy manually as usual.

🤖 Generated with Claude Code

[lakshvantb]

Superseded by #8: the workflow E2E run retried the original run's failed questions and recovered answers, so #8's rows come from strictly better data (this PR was generated from the older eval-branch snapshot).