SNOW-3010074 Fix incorrect closing of HTTP connections

[sfc-gh-pfus]

Overview

SNOW-3010074 Fixes a couple of problems related to HTTP connections closing.

1. Run a background thread, that monitors idle or expired connections and close them by setting evictIdleConnections and evictExpiredConnections. This one creates a thread that actively monitors such connections. This thread is closed on either JVM shutdown (daemon) or when HTTP client is explicitly closed.
2. Close HTTP client correctly. We cache HTTP clients, but in case of specific exceptions, we remove a connection from the cache. Added the code that closes HTTP client then, to prevent thread leaks.
3. Use setValidateAfterInactivity on a pool. With this setting, before retrieving a connection from a pool, the pool checks if the connection is still valid (i.e. it is not closed on the remote end).
4. Introduced net.snowflake.jdbc.idle_connection_timeout (default 30s) that is used as a timeout in all three mentioned functions.

[sfc-gh-snoonan]

In theory this looks fine to me. Couple questions:

1. Why keep the TTL argument at all? There's a constructor that doesn't require it.
2. The intent of the TTL is to handle DNS re-resolution and to keep load balancing of different IPs on the backend well distributed. Today, these change infrequently enough that it shouldn't be a huge deal, and we don't publish a ton of IPs. However, I'm a bit worried this will eventually lead to hotspotting. Arguably, we get our load balancing across multiple clients rather than from the connection pool for one client though. So I will call it ok. I'd like to see a test around the DNS update behavior though.

[sfc-gh-boler]

lgtm