Fix bug with windows version of mkstemp that causes failure to create more than 26 temp files

[DennisHeimbigner]

re: #1827

The issue is partly resolved by this PR. The proximate problem appears to be that the semantics of mkstemp in **nix is different than the semantics of _mktemp_s in Windows. I had thought they were the same but that is incorrect. The _mktemp_s function will only produce 26 different files and so the netcdf temp file code will fail after about that many iterations.

So to solve this, I created my own version of mkstemp for windows that uses a random number generator. This appears to solve the reported issue. I also added the testcase ncdap_test/test_manyurls but made it conditional on --enable-dap-long-tests because it is very slow.

I did note that the provided test program now fails after some 800 iterations with a libcurl error claiming it cannot resolve the host name. My belief is that the library is just running out of resources at this point: too many open curl handles or some such. I doubt if this failure is fixable.

So bottom line is that it is really important to do nc_close when you are finished with a file.

Misc. Other Changes:

1. I took the opportunity to clean up some bad string hacks in the code. Specifically
   • change all uses of strncat to strlcat
   • remove old string hacks: occoncat and occopycat
2. Add check to see if test.opendap.org is running and if not, then skip test
3. Make CYGWIN use TEMP environment variable