Timeouts specified in inconsistent units

According to our documentation, timeouts for MySQL and PostgreSQL are specified in seconds, while the same values for CQL are in milliseconds.

I think it's better to use milliseconds. The use case for this is that in HA scenarios, sometimes waiting for a second is too much and sub-second precision may be needed. Milliseconds is also the units used in HA.

The immediate reason why this popped up is NETCONF model definition. However, our current documentation probably should be improved as well. We currently have the parameters explained several times, once for each backend.

Parameters affected: connect-timeout, reconnect-wait-time, request-timeout. There may be others I missed.