Also add a safety check

The fix needs a followup: "ulong" is 32-bit on 64-bit Windows, so
xpand_connection_cursor::rowdata::length needs to be 64-bit but
this causes a cascade of datatype changes all over the code.

CLX-77 code used "thd->lex->result!=NULL" to check if the SELECT has
an INTO part. This is not correct, as the condition also holds when
we're using the PS protocol.
Use a more generic check: check whether the SELECT has any side effect.

Use the same solution as we used for MyRocks.

We don't need the MSVC hacks used by Spider.

Whoops, I did not mean to change the default of xpand_hosts.

Hardcode MYSQL_OPT_PROTOCOL to TCP in the Xpand connection. We can add support for sockets
later. Also fix a memory leak in xpand_hosts global variable.

Fix a number of issues that were preventing row-copying ALTER TABLE
from working:

- ha_xpand::create() and ha_xpand::open() should take the table name
  from the arguments, not from TABLE_SHARE object.

- ha_xpand::rename_table() should not use decode_file_path() as that
  fails with an error for temporary table names (#sql_nnnn)

- Maintenance of clustrix_table_oid through TABLE_SHARE::tabledef_version
  didn't cover the temporary work tables created by ALTER TABLE code.
  Switch to storing it in the Xpand_share instead, and update it in
  ::create() and ::open.

- On as-needed-basis, add quoting of table names to pieces of code that
  build SQL statements to be ran on the backend. This is a stop-gap
  measure until a real solution is implemented.

Fix the complaint from safemalloc about leaked 8 bytes.

When mysql_update() uses direct update, it will still create a quick select
(doesn't make much sense, does it?)
Quick select will be initialized, which will call ha_xpand::index_init
which will allocate ha_xpand::scan_fields.

Then, ha_xpand::index_end will be called but it will fail to free scan_fields.

Fix xpand.update test

It was not using the charsets correctly: UTF-8 encoded strings
were passed to MariaDB as latin1. The results of SELECT statements
looked correct because they were interpreted as UTF-8, too.

But the data in tables was mis-encoded so UPDATE statements already
didn't work correctly.

Direct UPDATE used to print the original statement. With Prepared Statement,
this could cause the statement with original parameter marks ('?') to be
printed.

Solve this in the same way as "Select Handler" does: print the statement back
from the parse tree. The parameters would be substituted.

Part #2: update the testcase as the database we're using
was changed.

Make Pushdown_select write output rows into select->join->result, instead of
thd->protocol.

This makes
- SELECT ... INTO @var
- SELECT ... INTO OUTFILE
- INSERT INTO myisam_table SELECT ... FROM clustrix_table
work as intended.

Also fixed the federatedx select pushdown handler:
- Do not fail an assert if the backend no resultset. Produce an error.
- For the SELECT .. INTO syntax, refuse to use Select Handler, because
the impelementation doesn't support this.