Improvements to ratio_split()

[damiendr]

The current implementation of ratio_split() distributes any change left after rounding to the last participant.

This causes two minor issues:

• participants with a ratio == 0 may still end up receiving change (debit or credit)
• the distribution of change is not fair, eg. one participant may end up paying 0.03 more instead of having three participants pay 0.01 more.

This PR fixes this by using the Largest Remainder method to allocate change between participants with a ratio != 0.

It also fixes a further problem whereby ratio_split() assumed DECIMAL_PLACES=2.

[coveralls]

Coverage decreased (-0.05%) to 95.433% when pulling 92423c9 on damiendr:better_ratio_split into 0ffcad1 on adamcharnock:master.

[adamcharnock]

Hi @damiendr! Thank you for this, and sorry for taking so long to get back to you. I've merged this now and it should be released soon.

[adamcharnock]

Released in 1.10.0, which also drops support for Python 3.5 and includes #40