Increase --batch_size for all jobs in main.yml

[sea-bass]

I measured these locally to fix the issue.

  ┌───────────────┬─────────┬──────────────────────┬──────────────────────┐
  │   Platform    │ Recipes │ Old batch size → KiB │ New batch size → KiB │
  ├───────────────┼─────────┼──────────────────────┼──────────────────────┤
  │ linux-64      │ 2269    │ 5 → 683 ❌           │ 10 → 402             │
  ├───────────────┼─────────┼──────────────────────┼──────────────────────┤
  │ linux-aarch64 │ 2269    │ 5 → 689 ❌           │ 10 → 406             │
  ├───────────────┼─────────┼──────────────────────┼──────────────────────┤
  │ osx-64        │ 2035    │ 5 → 595 ❌           │ 10 → 355             │
  ├───────────────┼─────────┼──────────────────────┼──────────────────────┤
  │ osx-arm64     │ 2035    │ 5 → 593 ❌           │ 10 → 355             │
  ├───────────────┼─────────┼──────────────────────┼──────────────────────┤
  │ win-64        │ 1691    │ 20 → 536 ❌          │ 25 → 439             │
  └───────────────┴─────────┴──────────────────────┴──────────────────────┘

But the root cause is that the testpr.yml that theoretically could have caught this had a hard-coded --batch_size 25 for all OSes, so it did not find this before merging the last PR.

[Tobias-Fischer]

Sorry I just did this on main ;)

[sea-bass]

thanks @Tobias-Fischer -- note that your numbers (all 25s) are more conservative on non-Windows so the jobs will be slower to complete... if that's how I understand this to work?