If the FM broadcast stations are further away, then putting it between the LNA and the RTLSDR may be OK, but if the LNA is bias-tee powered, the filter may block that power from reaching the LNA. In this situation you may not need the FM stop filter at all.
If you have serious issues with FM broadcast stations in your area, then put it between the LNA and the antenna. This will result in some degradation in system noise figure and hence system sensitivity, but it will stop the LNA and RTLSDR from overloading, degrading your system performance in other ways.
The LNA should be as close to the antenna feed-point as possible for best performance, and your RPi should be as far away from the antennas as practical to avoid your system being limited by noise emitted from the RPi, it’s power supply, and your ethernet cabling.
I know there’s a lot of stations out there with the RPi right next to the antenna, however I’m yet to see serious proof that these stations are not being limited by the noise from that setup…
Since my RPI3 is inside an aluminium case, I thought it would work as a Faraday cage, and thus would decrease the noise emitted by RPI3. Am I wrong?
Also in the picture bellow you can see my RTL-SDR + LNA. As far I know it is bias-tee powered, however I am not sure if I must power LNA with 5V from a external (without noise) power supply.
Putting in the RPi3 into a metal box can help, but noise can still escape down cables leaving the box. Just use a bit of extra coax.
The RTLSDR does look bias-tee powered. It also looks like the bias tee voltage input is going via a regulator, which might imply that it wants more than the 4.5v that the RTLSDR will provide. Best to check up on that.
The Bias-Tee output from the RTLSDR is filtered, and clean enough to be used for this kind of thing.