[PROPOSAL] Remove XDR dependency

[derobins]

PROBLEM

The venerable XDR library is used to perform I/O in the multi-file API calls. This is library is a problem for several reasons:

• It's very old (80s/90s) and has been removed from libc due to disuse
• It did not weather the 32-->64 bit transition very well. putlong and getlong were handled in different ways and have different type sizes on different platforms, which is annoying to ifdef around. In particular, the macOS system XDR is broken, requiring users to build the HDF4 XDR implementation.
• API quality is low (xdrproc_t, for example, has been fundamentally broken since the 90s)
• We only use a fraction of the XDR library's functionality
• Documentation is difficult to find
• Frequent source of bugs and issues in HDF4. This was significantly improved in 4.2.16, but it's still a problem.

PROPOSED FIX

XDR would be removed from HDF4 and replaced with new, significantly smaller, I/O code that provides similar, but more portable, behavior. This would be built directly into the HDF4 library and not as a separate, stand-alone library.

HOW WOULD USERS BE AFFECTED?

PRO
➕ HDF4 would be more portable and easier to build
➕ The HDF4 code would be easier to understand

CON
➖ There are calls in local_nc.h that take a pointer to an XDR struct and these calls (which should probably be in the internal API) would have their signatures change
➖ The next version of HDF4 would not be binary compatible with 4.2.16

NEUTRAL
✔️ All XDR-related configure/CMake options would be removed
✔️ The performance of HDF4 should be unaffected
✔️ The functionality of HDF4 would be unaffected

[derobins]

This will be addressed in 4.3.0 (#423).

[derobins]

Closing since XDR has been hidden in the mfhdf library in 4.3.0 and no symbols are externally visible. The remaining work is not high priority at this time.