Fixes #3336.

The blob type maps BLOB (and also TEXT) columns to PHP streams.
Internally, they use the ParameterType::LARGE_OBJECT (i. e. \PDO::PARAM_LOB)
binding type, which suggests that efficient handling of PHP stream resources
was intended.

However, at least when using the mysqli driver, stream resources passed into
insert() or update() are simply cast to strings. As a result, a literal
string like "Resource id #126" will end up in the database.

This PR fixes the issue by correctly processing streams in the
MysqliStatement when they are passed with the ParameterType::LARGE_OBJECT
binding type. It uses the mysqli::send_long_data() method to pass stream
data in chunks to the MySQL server, thus keeping the memory footprint low.

This method does not (despite claims to the contrary) allow to bypass the
max_allowed_package size!

The pdo_mysql driver was already capable of handling streams this way.
Now this is covered by tests.

Helpful documentation:

- http://php.net/manual/en/mysqli-stmt.send-long-data.php
- http://php.net/manual/en/mysqli-stmt.bind-param.php - see first "Note"
- http://php.net/manual/en/pdo.lobs.php
- https://blogs.oracle.com/oswald/phps-mysqli-extension:-storing-and-retrieving-blobs

Additional notes on MySQL's max_allowed_packet:

This change does not not intend to work around the max_allowed_packet setting,
and quick tests show that this is not possible: When MySQL is configured to use
a low max_allowed_packet value, an error will be triggered stating

  Parameter of prepared statement which is set through
  mysql_send_long_data() is longer than 'max_allowed_packet' bytes.

Documentation for the underlying mysql_stmt_send_long_data() C API function
suggests that max_allowed_packet is always a hard limit.

References:

- https://dev.mysql.com/doc/refman/8.0/en/mysql-stmt-send-long-data.html
- https://dev.mysql.com/doc/refman/8.0/en/server-system-variables.html#sysvar_max_allowed_packet
- https://bugs.mysql.com/bug.php?id=83958

What mysqli::send_long_data() seems to do is that every data chunk of data
passed to it is immediately sent out to the network. I have confirmed this
using tcpdump, and so the advantage might be that we can keep the memory
footprint low on the PHP side while processing streams.

1. Use the actual dsdriver distribution
2. Introduced `AbstractPlatform::getCharMaxLength()` since `CHAR` and `VARCHAR` limitations are different on DB2.

If a PRIMARY KEY field is defined in sqlite without autoincrement then
keys might be reused when rows are deleted, explicitly settings autoincrement
will prevent this.

Autoincrement is not enabled by default due to additional overhead introduced
with autoincrement bookkeeping and should only be enabled when the additional
uniqueness is required.

For testing purposes and not only, it may be needed to evaluate an arbitrary SQL expression (see https://github.com/doctrine/dbal/pull/3013, https://github.com/doctrine/dbal/pull/3108). Instead of doing `str_replace()` on an expected string, one could specify the expression explicitly.

The SQLite and Oracle platforms produces a `ROUND()`'ed or `TRUNC()`'ated difference between the datetimes. However it should be the difference between dates.

This should take care of most idiosyncracies.

See
https://stackoverflow.com/questions/1881853/select-without-a-from-clause-in-oracle

The Mysql family does not seem to like that one.