Following the previous PR to remove redundancy from _from_wkb_state
functions, do the same with _to_wkb_buf, using typed_value_to_wkb_buf.

Also, make static the endian_to_wkb_buf and bytes_to_wkb_buf functions,
as they are only used in this file, and remove the unneeded casts, to be
consistent with the corresponding functions in PostGIS.

text_to_wkb_buf was using bytes_to_wkb_buf, which swaps the byte order
according to the endianness. However, this is only needed for multi-byte
int or float types (and date/time types, which are large ints). Text
values are simply arrays of chars, so there should be no byte-swapping
going on. The initial int64 indicating the size of the string should
be swapped if needed, but this is correctly handled by int64_to_wkb_buf.

Also, we can avoid a call to text2cstring, which does an unnecessary
memory allocation. Use the VARSIZE_ANY_EXHDR and VARDATA macros to copy
from the text pointer directly.

Lastly, the old function was allocating space for the null terminator as
well, which is wrong, as we don't write the null terminator in the wkb
output. Fix this, and also correctly add the null terminator in
text_from_wkb_state. Palloc doesn't zero-out the memory, so this is
necessary or cstring2text could fail.

The error fixed in the previous commit was not detected as the tests for
the tbool, tint, tfloat, and ttext types did not test both the little-
and big-endian combinations of asBinary or asHexWKB. This commit fixes
this.