Use running averages to avoid overflows

[BWStearns]

The idea here is to use a running average instead of summing all the inputs. This adds an extra division operation for each element, but it has the benefit of not being able to overflow. It is currently failing for extremely large numbers.

[BWStearns]

I need to read up on how libraries should handle imprecision. I can't eyeball test_safe_circadian_with_big_numbers and know whether that difference is acceptable, but at least for test_with_many_reasonably_sized_numbers it seems like it's in the realm of acceptable.

---- tests::test_safe_circadian_with_big_numbers stdout ----
thread 'tests::test_safe_circadian_with_big_numbers' panicked at 'assertion failed: `(left == right)`
  left: `1.7014118e38`,
 right: `1.7014117e38`', src/lib.rs:191:9

---- tests::test_with_many_reasonably_sized_numbers stdout ----
thread 'tests::test_with_many_reasonably_sized_numbers' panicked at 'assertion failed: `(left == right)`
  left: `1499.6367`,
 right: `1500.0`', src/lib.rs:181:9


failures:
    tests::test_safe_circadian_with_big_numbers
    tests::test_with_many_reasonably_sized_numbers

[BWStearns]

Looks like the safe version works without a notable performance penalty.