Update Tests to be a little more Resilient

[ChaelCodes]

• Everyone gets generous epsilons, because tests based on time can be a bit flakey
• test_nested_collections is now influenced by TEST_SLEEP_SCALE

We should document TEST_SLEEP_SCALE somewhere.

[ChaelCodes]

For the record, I think epsilon is the right method. If we return 3 but it expected 200, then it'll throw an error - which we want, but it'll be generous with 200 vs 150 or 400 vs 260.

[jhawthorn]

Thanks!