[BUG] Cast String to TimeStamp issues

[tgravescs]

Describe the bug

We recently discovered that the config to turn off the string to timestamp cast was not being used. We fixed that with #705.

But while investigating that it seemed like a few things might be off. the format we are using and is used in CUDF is "%Y-%m-%dT%H:%M:%SZ%f". Based on what we support it seems like the ms section should be %SZ.%f. The pattern that we didn't match on but worked on the CPU was just: 2017-11-29 20:00:35. Ideally we would support this.

I think the regex's in this case are a pretty high performance overhead as well though so perhaps we should figure out different way to handle.

[tgravescs]

Note I have seen other formats for strings - such as 2020-09-07T01:05:57.840+0000
that we also don't support.

[sameerz]

This is related to issue #987

[andygrove]

This is also related to #1117 which I am currently working on

[andygrove]

Note I have seen other formats for strings - such as 2020-09-07T01:05:57.840+0000
that we also don't support.

@tgravescs The main issue raised in this issue (using the wrong cuDF timestamp formats) was resolved by #1718 and there is now a separate issue for reducing the regex overhead (#1738).

We do not support the format 2020-09-07T01:05:57.840+0000 and this is documented. Should we file a separate issue for adding support?

[tgravescs]

yes I think we should file one to say supported more formats

[andygrove]

I filed #1748 for supporting additional formats.