I've contacted the seller with the relay switch issue (wrong/non-switchable CH1 impedance). He was helpful but at the end I had to find it myself. He offered a discount (but nothing concrete) if I buy next time in his shop.
I have no idea how to contact BG7TBL.
Also as no input protection was mentioned in the spec, we can't complain about that and after all CH2 input impedance is also 50 Ohm but just around 6GHz...
Btw. I've tried to fix it again using the new information I've learned here (also updated the original post not to mislead people; through-hole resistor is used to have some inductance that cancels it out where the chips starts to have the correct input impedance). A photo attached. I know it's not the nicest job, but I did not want to stretch this anymore after several solders (first starting with SMD).
Note: If someone has access to 6GHz VNA or 6GHz SA + directional coupler + FA2 (not a too likely combination I know) I'd not mind knowing what is the correct lead length for proper crossover...
However if the question is if am I planning to directly complain, the answer is no. In fact if you are happy with relatively precise frequency measurements this device is great deal (especially if you get one that doesn't need too many fixes).
At this price point it's probably hard to find a better package for hobbiest use. It's also possible to build one, but it's even more time (I've calibrated FA2 with my own developed GPSDO so I'm quite aware of many challenges already, actually I'm still in search of an OCXO that has good phase noise despite I have already some double oven, SC cut and Japan made you name it version).
Sure, if you're willing to spend the money (or your job depends on that), there is no question that one has to settle a branded device (even if it cost significantly more) that meets its spec out of the box and you can have a proof of calibration (another cost factor that hobbiest are not necessarily willing to take).
Btw. talking about double oven, when using this device unboxed for a while I've noticed that its reference has shown a 10mHz offset (probably due to better airflow). Thus its temperature sensitivity would be another lengthy exploration...