Why miners try to produce big blocks?
Because transactions pay fees. The more transactions they include, the more fees can be collected. They prioritize transactions with higher feerate to maximize this.
What if, does a miner deliberately produce a block with a single transaction without considering the whole unconfirmed transactions, in order to gain the reduce of network latency advantage or some other purpose?
Some do that (though mostly because they just received a new block, and haven't had the time to construct a valid set of transactions on top).
In recent times, the argument for faster propagation also does not apply very much anymore. Techniques like Compact Blocks (BIP152), the Relay Network, and FIBRE can efficiently transmit blocks as long as they mostly consist of transactions that were seen before. This almost entirely removes the effect of size from the equation.