Charlie Zender - 2005-06-30

Hi,

I've committed a new change which reduces the number of critical
regions on reads from three to one.
The last one appears to be necessary, although I'm not sure why.
Let me know if you notice any funny (weird, not ha ha) regressions
as a result.

I've also begun introducing an opaque type, nco_char, to handle
NC_CHARs. There will be some compiler warnings for awhile but
regressions should still work.

Thanks,
Charlie