Relaxes the stiffness of the longest-first shuffle mode.

• The weighting scale is now `min(len(tests), 4.0 * sqrt(len(tests)))` — sub-linear, so prioritization no longer swamps the random component in large suites.
• `stdevs_above_mean` is clamped at `6`, so a single eternal test can't out-weigh every other moderately long test.

Adds a `shuffle_mode` expansion that a task can use to opt-in to a specific shuffle mode, typically for forcing pure random shuffling. Usage is like:
```

• func: "run tests"
  vars:
  shuffle_mode: random
  ```