Fix np.genfromtxt field name handling when dtype != None

[alanbriolat]

np.genfromtxt validates field names twice: once in genfromtxt and once in easy_dtype. Whilst the arguments to genfromtxt are used in the first validation, they aren't passed to easy_dtype (which is used only when dtype != None) and therefore in this case the default validation (strip non-alphanum, replace spaces) gets confusingly applied, ignoring genfromtxt's arguments.

This patch adds failing tests for the issue and fixes genfromtxt by passing the appropriate arguments onwards to easy_dtype.

This is probably the least invasive way to fix the issue. In my opinion, the whole thing is a nest of poorly-defined responsibilities between genfromtxt, easy_dtype and NameValidator, especially since the latter two are only used in the former. I'm willing to take the time to try and clean that up if it's likely to be well-received.

[jaimefrio]

It took me some time to realize that, if the current behavior was some form of intentional safeguard, it is very easily circumvented by simply setting dtype=None. If no one has any strong opposition, I would like to merge this, will give it a day or two for people to complain.

My only nitpick is whether the tests, rather than comparing against hardcoded values for the array and its dtype, should compare against the result of running np.genfromtxt on the same input, and with all arguments set identically, except for dtype=None.

[charris]

Squashed commits and rewrote commit message in #5459. Thanks @alanbriolat..

[charris]

As you say, genfromtxt is a bit of a mess. Any work you want to do cleaning it up is welcome.