functions: type autoinference

[IvanGrigorik]

This PR addresses #342.

Fixed type inference for functions, so now no need to specify argument types.

[IvanGrigorik]

@kennykos can you please check and confirm that it fixes #342?

You have another problem with this line.
Pykokkos requires specifying the types of new values, since autotype inference for all values can lead to some problems (we can fix it in the future, but this PR addresses only the argument inference part.

[kennykos]

@kennykos can you please check and confirm that it fixes #342?

You have another problem with this line. Pykokkos requires specifying the types of new values, since autotype inference for all values can lead to some problems (we can fix it in the future, but this PR addresses only the argument inference part.

This fixes #342, given my caveat above about the mangled error message when the return type is not specified.

[kennykos]

@IvanGrigorik the most recent push checks argument inference for nested @pk.functions; the tests fail.

Is this something we can support as well.

[IvanGrigorik]

@kennykos everything is fixed
There was some problems with paths and default arguments like wid, acc, etc

[kennykos]

I have some questions

1. Why are we explicitly casting common thread ID names to int?
2. Why are we casting ints to doubles?
3. Should we consider other module types in addition to numpy?

See the specific comments for details.

Otherwise, this looks good to me.

[IvanGrigorik]

For now pykokkos View type is supported, but eventually we will deprecate them as well

[kennykos]

Can we have some sort of failure warning here if the iterations max out and there are still uninferred types?

[kennykos]

Looks good to me! I left one comment about an error message that could be nice, but I'll defer to @IvanGrigorik on that one. Otherwise, looks good to me!

[kennykos]

Looks great!