Fix two problems:
- The value of mysqli_get_client_info() has been changed recently and did
not include "mysqlnd" anymore thus the test suite was thinking the build
is always libmysql. This did not kept the suite from running pconn tests
- Going back to the libc allocator because the memory arena could be on a
persistent connections. If the build is not debug there will be no error
but the memory will be freed and in the second use of this pconn freed
memory will be used - not good! For now the arena doesn't take an argument
whether it should allocate persistently or not, thus persistent is safe
for now.
Johannes gave his +1 to commit this.
Fix a bug with mysqlnd_fetch_field(_direct()). With mysqlnd the optimised
function was called, which however, doesn't respect that during store the
raw data is not unpacked, to be lazy. The data is unpacked to zvals later,
during every row fetch. However, this way max_length won't be calculated
correctly. So, if a mysqlnd_fetch_field(_direct) call comes we need to
unpack everything and then calculate max_length...and that is expensive,
defies our lazy unpacking optimisation.
Clearly separated fetching (physical reading) from decoding phases (data
interpretation). Threaded fetching added but disabled as needs more work for
Windows. For Linux needs some touches to add pthreads if this is enabled,
probably with a compile-time switch.
The code reorganisation makes it easy to add also async API, similar to
cURL's one.
- Clean up a bit a type mess - 4 types less. No need to have aliases
for the same thing (unification is the name).
- New macro for Johannes mysqlnd_stmt_ro_result_metadata
