Rearrange thrust includes for CUDA 12.5 compat

Address Issue #119

ExampleProvider now has save_mem_caches and load_mem_caches so the
current state of the in-memory cache can be saved and restored.

Version bump

version bump

Better multi-device support

Provide helper functions for setting current device.
Throw useful error message if tensor is not on current device.

0.5 release

Support for linear grid interpolation (forward only for now).
Better upport for pytorch data loaders.
Various bug fixes and documentation tweaks.

Fixed bad number in test

improved test case

Fix for using custom typers and molcaches

The python bindings were providing empty molcaches when explicit typers
were provided to ExampleProvider.  Now the extractor will get molcaches
from the settings, replicating the lig molcache as needed to fill out
the typers.

cmake refinements