RD-15247: Reimplement JQL generation

[bgaidioz]

Main issue is that we do not rewrite IS NULL or IS NOT NULL correctly. They come as x = NULL or x != NULL. In JQL they have to be rewritten as IS EMPTY or IS NOT EMPTY. That's what is fixed here.

It's now possible to filter issues without a description (or with a description with IS NOT NULL).

SELECT components          
FROM jira.jira_issue
WHERE description IS NULL
LIMIT 10;

Still, some fields can't be filtered with IS NULL, it's enforced by JQL. Like that field id used in the original bug report.

Later the ticket was changed to failing on epic_key IS NULL. epic_key can legitimately be null, but filtering it for = or != can't use the same keyword as when filtering for IS NULL.

This patch is refactoring the JQL generation, case by case. That has some advantages:

• supporting IN and NOT IN when supported (this is new), or IS NULL
• skip predicates that can't be implemented (e.g. self = ...) while before we were assuming all were supported,
• track whether or not all predicates were applied, which will eventually permit to fix RD-15251 (pushing of LIMIT) or tell the user,

I had to delete two tests that were testing the infrastructure, since now the DasJiraTableManager needs connectivity to a JIRA server to know the user timezone.