Allow to find capacity over multiple regions #535
Description
Currently, dstack supports configuring multiple compute regions only for LambdaLabs. Users should be able to set up multi-region compute for all backends to increase the chances of GPU availability (especially relevant for spot).
Clouds do not provide the information on spot instance availability per region. So what we can do is to let users specify a primary compute region (just like now) and any number of additional compute regions. Then dstack run will try to use the primary region first and then additional regions until a provisioning attempt succeeds.
We can use different heuristics to determine the order of regions. For example, AWS provides Spot placement score. For Azure, spot demand and availability can be guessed from pricing history.