Do you want to measure the Raw PWM induced noise, or the average current at that given moment, as that will influence how I approach this,
First up, I would group the current sense and buffer as close to the current shunt as within reason, the smaller your loop areas the harder it is to pick up noise
If you only want an averaged current, I would be filtering it down with some capacitance at the current sense amplifier (after its series resistors), to trim it to what the amplifier can produce, then possibly at its output to lower to what the ADC can measure
That area of the board will be getting quite hot, so I would have them floating on there own little Via enclosed island, to keep the temperature gradient pretty low across the CS and buffer,
I'm assuming you don't really care about the lowest of the the low currents, as with no offset from ground you will likely see issues with low currents where the amplifier and ADC don't really have any means to quickly drag the signal that close to ground,
Now that the current signal is filtered to what you can measure, and buffered to a low output impedance, your more free to route it furthur away from the current shunt, I would not go too crazy, but a few cm's will be OK, just treat the buffer ground as the negative of a differential signal and don't join it to your main ground plane at the ADC, As for fancy ground planes, Have the branch close to the ADC, then run to the CS/Buffer run your signal / return for the signals over the plane connection between them,
For the FGPA / Bluetooth stuff, spinkle decoupling caps like the datasheet recommends, place them as close as you can to there respective power pin groupings, the back side of the PCB is always an option, and for your ADC-FPGA data lines, you may want to throw some 50-220 ohm resistors on the output for that data line to reduce the bandwidth of the radiated signal edges, your near something really noisy, but no need to make more noise.
And finally, data lines always over an unbroken ground plane, if you need to hop with a via to another layer, slap a via to ground close by, so the return current can follow it, (path of least impedance)