Getting the count of objects is very slow in large PostgreSQL databases

[jamadden]

We use the generic query SELECT COUNT(*) FROM <table>, but that uses a full table scan which can be very slow. In a history-free database, <table> is object_state, which contains large rows and this query can take many minutes to run (on my system, for 400,000 rows with a slow-ish disk, it takes 342.794s (5 minutes)).

This is probably not a bottleneck for most applications, but it does hinder certain zodbshootout operations (specifically, when --min-objects is used). Can we do better?