Balance CI for testing

[willtebbutt]

There are currently loads of CI runners, and we can almost certainly get away with many fewer. This reorganised CI such that

1. the longest run should be slightly shorter, and
2. there are fewer runners used.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

[github-actions]

Performance Ratio:
Ratio of time to compute gradient and time to compute function.
Warning: results are very approximate! See here for more context.

┌────────────────────────────┬────────┬─────────┬─────────────┬─────────┐
│                      Label │  Tapir │  Zygote │ ReverseDiff │  Enzyme │
│                     String │ String │  String │      String │  String │
├────────────────────────────┼────────┼─────────┼─────────────┼─────────┤
│                   sum_1000 │  105.0 │     1.0 │        5.41 │     1.9 │
│                  _sum_1000 │   7.89 │  1430.0 │        33.0 │   0.106 │
│               sum_sin_1000 │   2.46 │    1.61 │        11.0 │   0.988 │
│              _sum_sin_1000 │   3.41 │   341.0 │        18.6 │    1.54 │
│                   kron_sum │   59.7 │    9.93 │       196.0 │    13.5 │
│              kron_view_sum │   70.0 │    9.74 │       185.0 │     9.1 │
│      naive_map_sin_cos_exp │   3.19 │ missing │        9.05 │    2.83 │
│            map_sin_cos_exp │   4.72 │    1.76 │        7.65 │    3.47 │
│      broadcast_sin_cos_exp │   4.33 │    2.58 │        1.69 │    2.86 │
│                 simple_mlp │   8.81 │    3.45 │        13.4 │    3.35 │
│                     gp_lml │   14.6 │    4.53 │     missing │ missing │
│ turing_broadcast_benchmark │   7.37 │ missing │        34.8 │ missing │
│         large_single_block │   3.88 │  4380.0 │        31.9 │    2.24 │
└────────────────────────────┴────────┴─────────┴─────────────┴─────────┘