Use larger base type for TDFASet + small refactoring.

This does the #39508 (closed) and has zero effect on speed (which is strange, considering that set of byte is observably faster and that my tdfasets reach length 12 with array of bytes and only 2 with 64-bit array of PtrUints), but looks like a common sense to me. Really, the only reason to use byte bitfields I can imagine is being able to serialize/deserialize them to/from a binary blob with a single Write/Read, and even then...

Also this gets rid of some global variables used to communicate with callbacks in favor of local variables and opaque arguments of these callbacks, and uses a very clever trick to get rid of the global variable SearchNodeProcPointer.