Unify datetime parsing

Various places in pandas convert a series of datetime strings to datetime64/timestamps/etc' objects. read_csv, to_datetime, upcoming
sql improvements and probably elsewhere.

The code paths each takes are seperate so that fast paths aren't consistently available, and it's difficult to apply improvements such as #6021 in one place
and have them apply everywhere. and there's room for internal incosnistencies in parsing behaviour (#5958).

https://github.com/pydata/pandas/pull/5950/files#r9007497
#4826
#5993

Do something about it.

cc @jreback , @danbirken

[jreback]

pretty much closed by #6021, #5490