Handle pg_runtime_variable return values

[marcusmartins]

When querying postgres pg_runtime_variable, postgres might return -1
for max_standby_archive_delay and max_standby_streaming_delay

[wrouesnel]

Should this perhaps be a math.Inf(1) for these values since they both technically imply infinite wait times, and Prometheus supports that representation?

[wrouesnel]

So I let this languish far too long, but the answer is kind of obvious based on my prior experience - math.Inf(1) in Prometheus tends to break graphs and in this particular case the lead up to a -1 is not steadily increasing wait times AFAIK. Merging.