Use a serde visitor in inspect to avoid buffering the full database

[althonos]

Hi Jim!

In light with the inspect release here's a PR to reduce the memory requirements: using a serde::de::Visitor to visit the database allows to accumulate the stats over the GenomeSketch without having to deserialize the whole Vec<GenomeSketch> in memory. This means now you can sylph inspect gtdb-r220-c200-dbv1.syldb in about the same time but 60MiB of memory. An example for this kind of things can be found here in the serde documentation to process a large JSON file without buffering it first.

NB: I was exploring how to do that with the database during the search as well to avoid having to load the full database, so that it could run on more memory-starved systems. What I had so far worked okay for single threads but didn't parallelize well, however that may be a direction worth exploring.

[bluenote-1577]

Hi Martin, this looks brilliant. Thanks so much for your help, as always! I really need to learn more about serde :). Tested and looks good to me.

I thought about not loading the full database for sylph as well, which could lead to better memory costs. Ultimately, for the forseeable future, I don't see dereplicated databases getting too much larger. So the database will be < 30 GB which I think is doable for most folks.