Fix provider status tracking bugs

[Redm4x]

• Add timeout to grpc status calls (10s, same as rest calls)
• Cap available resources to 0 when allocated is larger than allocatable to prevent negative numbers. This can happen due to provider overcommit.
• Fix a discrepancy between provider capacity tiles and their corresponding graphs. This was caused by an incorrect handling of provider snapshots when status endpoint is not accessible.
• Fetch status of recently online providers first (so that they come back online sooner after a temporary outage)

[baktun14]

is it supposed to be bytes?

[Redm4x]

It's fine to keep it as a string I think. Only use for a numerical type would be to query GPUs with X vram "or more", but for now it's simpler like this. I removed the todo, we can always change the type in the futur if needed.