Tackles #505, and maybe fixes #483.

This PR adds tests and short-term fixes for a myriad of downsampling problems, largely caused by sourmash-bio/sourmash#3384, in which Collection:sig_from_record does not downsample the loaded sketch.

Specifically this PR:

• fixes fastmultigather to downsample the query properly;
• fixes fasmultigather on a RocksDB to properly handle scaled, includig downsampling the query, and also checking the max scaled on the RocksDB;
• fixes manysearch to downsample both the queries (via MultiCollection) and database sketches, as well as output scaled, ksize, and moltype;
• fixes manysearch against RocksDB to downsample the query, as well as check the max scaled on the RocksDB and output scaled, ksize, and moltype;
• fixes multisearch and pairwise to downsample queries and against, as well as output scaled, ksize, and moltype;
• adjusts MultiCollection::load_sketches to downsample them to the provided selection (as used in manysearch;
• adds tests for the above.

I'm doing a patchfix release with this because it's returning incorrect results ;(. At some point we will want to revisit the code and clean it up more - see #510.