Ensure we are resilient across PG versions

[sgrif]

We currently always use the binary format for PG. However, the only type which explicitly documents its binary format is integer types, which is simply documented as using network byte order. For all other types we literally have to read the PG source code to find the binary representations. This appears to have been reasonably stable for at least the last couple versions, but we shouldn't assume that will always be the case. In fact the documentation explicitly states:

Keep in mind that binary representations for complex data types might change across server versions; the text format is usually the more portable choice.

I do want to continue using binary whenever possible. For some types, the difference in performance is several orders of magnitude in terms of the work being done to handle the binary form vs what we'd have to do to parse the text form. However, we should at minimum do the following:

• Test against as many PG versions as we can and is reasonable
• Check the server's version in the connection, error if it's too old, and at least warn if it's newer than the most recent version we've tested against.
• Potentially fall back to the text form if the version isn't one that we know uses the expected binary format