Data insertion optimization

[nsmith22]

Update the target-postgres dependency to get the latest master branch commit, which is many commits ahead of the 0.2.4 package version. This provides several performance optimizations that have been added, such as:

• Multiple optimizations: Multiple performance improvements target-postgres#204
• Reduce state message memory footprint: Reduce memory footprint of state messages target-postgres#212

Additionally, bypass table insertion work when the record count is zero. The code still respects the persist_empty_tables setting to manage the table schema itself, but will not go through the expensive process to perform a zero-record insertion. That process takes ~6s per table and is effectively a no-op. For table-heavy taps, this saves a lot of wasted time.

Testing

Tested locally with a few of my own taps. A tap with ~25 tables and no new data was able to run in ~40 instead of ~1m45s, mostly due to the code which skips table data insertion when the data batch has zero records.