Changes for numpy 2.0 to fix test matrix.

[Thrameos]

This PR deals with the broken test matrix following the removal for float_. I replaced it with float16 though that is actually a different test. I dealt with the required alterations to support half type for array conversion, though I can't find a good pattern to deal with byte reversed arrays so that may be untested.

[codecov]

Codecov Report

Attention: Patch coverage is 63.82979% with 17 lines in your changes missing coverage. Please review.

Project coverage is 87.33%. Comparing base (cdb6be2) to head (f6849c5).
Report is 160 commits behind head on master.

Files with missing lines	Patch %	Lines
native/common/jp_convert.cpp	63.82%	15 Missing and 2 partials ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1196      +/-   ##
==========================================
- Coverage   87.44%   87.33%   -0.12%     
==========================================
  Files         113      113              
  Lines       10238    10281      +43     
  Branches     4059     4065       +6     
==========================================
+ Hits         8953     8979      +26     
- Misses        691      706      +15     
- Partials      594      596       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Thrameos]

Seem like I gave one more set of changes before this is complete. Not sure why onle one version triggered it.

[Thrameos]

I am unable to replace the error seen in the test suite but it most likely is a Python behavior not a JPype one. The test was making sure that all cast types go through required paths. Python itself through the exception that "e" type is not a simple data type. If they choose to change that behavior then we shouldn't test against it.

@marscher I believe this is ready to go.

[Thrameos]

Not much to this other than changes to the test bench to get it back on line.

The only odd code was to better support the half float which is valid buffer type in Python. There is no native Java type for it.

[Thrameos]

This is a crude implementation of the "half" floating type. There is no native type in Java to support this but it is possible that someone could pass a half floating in for promotion to either float or double. This code is exercised in the test suite.

[Thrameos]

I am not sure if there is every a case in which a reversed byte order half floating value would be converted. Through the order of this operation is reasonable to define. Reverse the bytes, promote to float then convert to Java type.

[marscher]

Thanks. LGTM