Correct PostgreSQL metrics for dead tuples, with integration tests and refactoring

[easimon]

Larger version of #2468

• Corrected PostgreSQL metrics for dead tuples
• Refactored PostgreSQL metrics, extracted String constants for less repetitions.
• Added integration tests to verify (esp.) the dead rows gauge.

1. pg_stat_user_tables has a row for each table in the database, so selecting the first row of it results in "dead rows for a single random table". Calculating the sum over n_dead_tup results in all dead rows instead.

2. pg_stat_database has only one row for each database, so the sum() over numbackends was harmless but unnecessary, so I removed that one.

3. At least for PostgreSQL, there's a difference between getObject(n, Long.class) and getLong(n). The former can only convert INT4 and INT8 (but not NUMERIC and DECIMAL) to Long, the latter converts all numeric types. sum(n_dead_tup) is such a case where this matters, since its type is NUMERIC.

I could understand if someone finds this too much of a change, so I created both MRs.

[shakuzen]

Looks good to me. Thank you for finding the issue, sending a fix, and adding integration tests for us. Much appreciated. I will rebase the changes on our 1.7.x branch so that we can include it in our next round of patch releases.

[shakuzen]

Sorry for the long wait to get this in. Thank you.