The idea of connecting the current sense resistor directly to the PIC
input is neat.  I wasn't aware you could run the PIC's reference voltage
down to 0.2V without severe accuracy problems.  Unfortunately, you can
only measure one direction this way, and the application is bipolar.
With two more resistors you could add some DC bias that would allow you
to measure bipolar current.

The commercial current sensor is nice, but you want ten times the
sensitivity that it has.  Can you get a more sensitive one?  Adding a
gain of 10 to it is not hard, but getting the correct DC bias (reference)
might be.

Does the sensor have a reference output?  If so, just build an inverting
amp referenced to that, and you're set (one op-amp, two resistors,
compensate for the inversion in firmware).  If not, you have to generate
one, and that can get tricky.  Is the reference really 2.5V, or is it a
fraction of the power supply?  If your power supply voltage is not
exactly perfect (both accurate and stable), that makes a big difference.
A 2.5V reference can be generated easily, using an LM4040.  Be careful of
tolerances, though.  a 1% error in your reference voltage or a 25 mV
offset voltage in your op-amp looks like 1 amp to your firmware!

Most of the errors can be calibrated out, but temperature drift can be a
problem.  Make sure your power supply, sensor, amplifier circuit, and PIC
are at constant temperature (indoors) and use only one supply if
possible, with short wires.

If the commercial sensor has a reference output, that may be the most
accurate solution, but my favorite of all the solutions is the
roll-your-own, using an op-amp and a handful of resistors.  I think
you'll get good accuracy and stability that way, and it's a lot cheaper
than the commercial sensor.

Don