Resolve Windows access violations in JDBCBackend

[khaeru]

This issue collects information from iiasa/message_ix#452 (comment), iiasa/message_ix#453, and pytest-dev/pytest#8510. These seem to be related to #229, #247, and jpype-project/jpype#561.

Briefly:

• Something in the Java code behind JDBCBackend (ixmp_source) triggers various errors, including "Windows fatal exception: access violation."
• These seem to cause no harm (see descriptions at “Windows fatal exception: access violation” on Windows using pytest jpype-project/jpype#561), except they create a lot of console noise (better error message when jdk not installed #249). Suppress Java noise on Windows/AppVeyor #247 did not resolve these, but introduced a workaround to silence them.
• However, more recently (while working on Install GAMS using iiasa/actions for CI message_ix#452) Windows CI jobs began to fail because pytest returns an exit code 127. This is not an exit code that pytest should ever return.
• One hypothesis (Pytest gives exit code 127 pytest-dev/pytest#8510) is that something is failing during the process of shutting down the Python interpreter, after pytest has successfully completed. In other words, pytest completes and is ready to return an exit code 0, but this inscrutable exception (probably in ixmp_source) causes (via JPype) an exit code of 127 to be returned, instead.

Status:

• After about ~1.5 days of work, it was not possible to isolate the issue.
• As an interim measure, Work around Windows CI failures message_ix#453 introduced a workaround for this specific case.

A true fix will be to address the flaws in the Java code; this should get rid of both the console noise (#229) and obviate the workaround in iiasa/message_ix#453.