Investigate Blob insert performance

[ecton]

On the Scaleway VM, Nebari's inserts land inbetween Sled and SQLite in raw throughput, but locally on my machine, Sled and SQLite are able to maximize my NVME storage -- 1.5GB/s while Nebari caps out at around 430MB/s.

It would be interesting to profile large blob insertions and see what the performance bottleneck is. It might also speed up other things.