Optimize getFilesAndIndicesForDimensions

[mgrunbauer]

This MR makes the following changes:

• When requesting data with multiple dimensions, instead of executing queries per dimension, execute a single query for all dimensions. This is happening in getFilesAndIndicesForDimensions and getTableNameForPathFilterAndDimensions.
• Check if pathfiltertablelookup table exists only once during adaguc-server lifetime, instead of once per table access.
• SQLite will now return dim values in the same format than PSQL (e.g. always 10.0 instead of 10)
• You can now run all tests against the postgresql backend.
• The maxquerylimit limit now limits the total results from getFilesAndIndicesForDimensions() instead of limiting on the first dimension queried. This fixes: Simplify query used by CDBAdapterPostgreSQL::getFilesAndIndicesForDimensions #341

All changes combined result in less queries. For example GetMap required 12 queries before, only 3 queries with this MR.

This slightly improves performance, running the complete benchmark locally takes 140 seconds on master, 128 seconds on this branch.

I've replaced some of the getTableNameForPathFilterAndDimension() calls with getTableNameForPathFilterAndDimensions() (plural dimensions). The former is now a wrapper function that executes the latter, so the code path is identical. Replacing the remaining getTableNameForPathFilterAndDimension calls requires some refactoring.

---

Things out of scope for this PR:

• SQLite will not implement the changes made to the PostgreSQL class. SQLite might become deprecated.
• Fix test dependencies when running tests using PostgreSQL. I worked around this with a "temporary band aid" (drop all tables).
• "query hidden" with SHOW_QUERYINFO is not required for now.
• Postgres is still using real to store floats, which will result in precision loss. We should probably use double precision. I've decided to skip it in this MR for now.

[maartenplieger]

Nice!!!!! I am deploying a dev version for geoweb to spot any issues.