Electronic – chip antenna exposing U.FL male

All WiFi systems I've seen use a 50 ohm impedance and apart from that you want the cable with the lowest attenuation at 2.4GHz. Wikipedia has a fairly nice table to get you started: Coaxial cable standards. The table lists the attenuation at 750MHz but the comments area has attenuation at 2.4GHz for many types of cable. Some entries aren't complete but it'll give you the most common types so you can find datasheets where you can check the atennuation / loss at 2.4GHz.

The other thing to look for is the size and cost of the cable. For example RG-58/U has an outer diameter of 5mm, suitable SMA connectors and the cable are widely available at reasonable cost but attenuation is shown as 1.056 dB/m @ 2.4 GHz. The 6-7dB loss for your length of cable is quite high and will probably halve the range.

LMR-1700 coax has a loss of 0.056 dB/meter @ 2.4 GHz however its outer diameter is a rather large 42mm making it hard to terminate properly onto SMA connectors. Maybe something like LMR-195 would be a good compromise.

Another thing you could do is use a larger / lower-loss cable with a pair on N connectors at each end and a pair of N to SMA adapters, I've seen many commercial installations that use that technique. If you go that way be sure to buy good quality adapters that specify the insertion loss at 2.4GHz.

The other and probably the best technique if it's possible in your case would be to extend the data and power connections to the access point so it can be placed nearby the antenna. You may find that an IP67 enclosure so you can put it all outside is actualy cheaper than the higher grades of cable and connectors.

From the datasheet it always transmits over one of the antenna connectors and a software algorithm is used to determine the best one at the time to use for receiving:

BCM4312 has 2 radio chains, one is a dedicated Rx chain, the other is shared by TX and RX chain using an RF switch. For RX, It supports antenna diversity and provides four options. A software diversity algorithm is used to determine which option is used.

It doesn't mention what the four options are but in general the idea of using diversity mode is when you think the reception from one antenna is likely to be better than the other, it won't increase range otherwise because only one is actually being used at a time. For further reading here's an article Multipath and Diversity from Cisco that goes into quite a bit of detail on the use cases.

So it's certainly possible to connect a Yagi to each antenna connector and it shouldn't cause any problems as long as you don't use it to try and expand coverage to a wider area. But if it's in a location where multipathing isn't likely to be issue like say outdoors and line-of-sight it won't gain you anything either. In that case you'd be better to spend the extra money on a better antenna and low-loss coax and connectors.

From a practical point of view determining the best setup is mainly a matter of testing various configurations and doing a site survey to make sure all the areas you're interested in have good coverage. The last time I did something similar it was an embedded system that had logging and GPS anyway, but there are a number of phone apps around that may prove useful if the clients will be moving around. For point-to-point you could just measure throughput and reliability over a period of time to make sure it meets your needs.