Arrow: Refactor humanize() method, expand usage, fix edge cases

[systemcatch]

.humanize() is getting kind of long and could do with refactoring. We'd also like to explore allowing other inputs such as date and timedelta objects.

[Songyu-Wang]

Hello,
I am bored at home so I decide to take a look at this issue.
After examining the code,
I feel the humanize() is flawed.

import arrow

```python
x =arrow.now()
y =x.shift(seconds=46)

```python
x.humanize(y)

'a minute ago'

x.humanize(y,granularity='minute')

'0 minutes ago'

Specify granularity should not cause an inconsistent result

[Songyu-Wang]

I feel a better approach is to do something like this:

true_granularity is the granularity
if granularity=auto:
  figure out and update true_granularity
do the calculation with true_granularity
handle the list granularities

[systemcatch]

After we fix #848 we're left with a few edge cases.

Months still present a problem.

(arrow) chris@ThinkPad:~/arrow$ python
Python 3.8.3 (default, Jul  7 2020, 18:57:36) 
[GCC 9.3.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import arrow
>>> dt=arrow.utcnow()
>>> dt
<Arrow [2020-11-08T15:12:18.911919+00:00]>
>>> later=dt.shift(months=+1)
>>> later
<Arrow [2020-12-08T15:12:18.911919+00:00]>
>>> dt.humanize(later)
'4 weeks ago'
>>> later.humanize(dt)
'in 4 weeks'

Also leap years will need dealing with.

[yiransii]

Hi, I can take a look at fixing the edge cases.

Is it supposed to print "1 month ago" and "in 1 month" for the month+=1 example?

@systemcatch

[systemcatch]

@yiransii yes that is what we would expect to happen here.

[yiransii]

@systemcatch
So the edge case happens because a month might have 30 or 31 days, but self._SECS_PER_MONTH: is always set to the value of 31 days in the condition elif diff < self._SECS_PER_MONTH:

In the above month+=1 example, dt=arrow.get("2020-11-08T15:12:18.911919+00:00") returns 4 weeks cause Nov has 30 days, which is smaller than 31 days, so it enters the if statement and uses weeks to describe the delta.

Pull request submitted! I have also added test cases to test humanize with the month+=1 delta for each month.
https://github.com/arrow-py/arrow/pull/894

[jadchaar]

Once this is fixed, we should revert coverage requirement to 100%: https://github.com/arrow-py/arrow/issues/749.

[anishnya]

What level of accuracy are we seeking? I can working on a solution with similar accuracy to https://www.timeanddate.com/date/duration.html, however, this may not necessarily be the most efficient solution.