I did a tracker board for my RF module last year and got the same « slow fix, half range » result on rev A, so this is mostly what my lecturer took apart on it.
Keep the top pour, but make it a real ground: connected to the layer 2 plane with stitching vias everywhere, especially along every RF trace and around each RF section. Space them about 3 to 5 mm apart, tighter near the RF parts. A pour with a few random vias is a floating patch that resonates, and that alone can cost you sensitivity.
Do not split the ground. Isolate the sections by placement (each radio in its own corner, the noisy LTE power stuff away from GNSS) and, if needed, by a stitched fence of vias, but keep one continuous plane underneath.
Under and around the chip antenna and the patch: follow each antenna’s datasheet keep-out exactly. No copper on any layer in the chip antenna’s keep-out, and the patch wants its specified ground area under it. Michael is right too: give the two antennas different edges and ideally opposite corners.
Run your RF traces as grounded coplanar waveguide at 50 ohms (calculate with your stack-up, the vias on both sides are part of it) and leave a pi-network footprint at each antenna feed so you can tune later.
Then measure: send the LoRa RSSI and SNR and the GNSS satellite count and time to fix as readings with every message, and compare rev A against rev B on one chart. That’s the easiest A/B test you’ll get.