frequencies: Unhandled error with missing or invalid dates

### Current Behavior `augur frequencies` crashes with an unhandled error when any used sequence has a missing or invalid date (e.g. empty string, `XXXX-XX-XX`, or other unrecognized values that `get_numerical_date_from_value()` returns `None` for). Note: unused sequences (not in the tree/alignment) do not trigger the error, since their dates are never accessed. ### Expected behavior A clean error or no error. ### How to reproduce The exact error depends on the mix of date values. `get_numerical_dates()` returns a dict with float values for exact dates, tuple values for ambiguous dates (or date ranges), and `None` for invalid/missing dates. 1. When the dict contains tuples, the underlying array is object-typed and `None` stays as `None`, causing a `TypeError` in `np.mean()`. 2. When all valid dates are exact (floats only), pandas coerces `None` to `NaN`, which passes through `np.mean()` but later causes a `ValueError`. <details> <summary> Scenario 1: with ambiguous dates </summary> ```sh cat >tree.nwk <<~~ (A:0.1,B:0.1,C:0.1); ~~ cat >metadata.tsv <<~~ strain date A invalid B 2016-01-06 C 2016-04-XX ~~ augur frequencies \ --metadata metadata.tsv \ --tree tree.nwk \ --method kde \ --pivot-interval 3 \ --output out.json ``` Output: ``` File ".../numpy/core/_methods.py", line 131, in _mean ret = ret / rcount TypeError: unsupported operand type(s) for /: 'NoneType' and 'int' ``` </details> <details> <summary> Scenario 2: with exact dates </summary> ```sh cat >tree.nwk <<~~ (A:0.1,B:0.1); ~~ cat >metadata.tsv <<~~ strain date A invalid B 2016-01-06 ~~ augur frequencies \ --metadata metadata.tsv \ --tree tree.nwk \ --method kde \ --pivot-interval 3 \ --output out.json ``` Output: ``` File ".../augur/frequency_estimators.py", line 840, in float_to_datestring days_in_year = 366 if isleap(int(numdate)) else 365 ValueError: cannot convert float NaN to integer ``` </details> ### Possible solutions 1. Error on missing dates in the tree or alignment. Example from [nextstrain/WNV#118](https://github.com/nextstrain/WNV/issues/118): ``` ERROR: The following sequence ids are missing valid dates: 'AF202541' has date 'XXXX-XX-XX' 'AF206518' has date 'XXXX-XX-XX' 'AF260968' has date 'XXXX-XX-XX' 'AF481864' has date 'XXXX-XX-XX' ``` 2. Error on **all** sequences with missing dates, regardless of whether they are in the tree or alignment. - This is a stricter data quality check that's probably not necessary. 3. Automatically drop sequences with missing dates. - This could hide underlying data issues and produce unexpected results. --- First noticed in https://github.com/nextstrain/augur/pull/1913#discussion_r2453394839