There are no set recommendations as this is one that is entirely up to personal preference.
For DSD the difference between filters 1 and 4 is how much out-of-band (ultrasonic) information is removed prior to decoding. F1 has the widest bandwidth and F4 has a sharp cutoff close to the audio band. In most systems I see F1 and F2 used and in my own system I usually bounce between the two. F4 is there primarily to handle the situation of amp and speaker combinations that don’t get along well with lots of ultrasonic information.
If you’re upsampling to DXD then PCM filters 1 - 4 are a variation on the traditional finite impulse response filter that’s been in use for ages. The differences are related to rejection of Nyquist images (noise) vs transient response. The basic trade-off is that the tighter the filter the more relaxed transients will be. F5 is a Gaussian response filter which is linear phase with no overshoots on transients (i.e. a very well-behaved linear filter). F6 is asymmetrical with non-linear phase response (also called minimum phase). The benefit here is that there is no pre-ringing in the filter, but the trade-off is that there is more ringing after the impulse (which some would argue is more natural).
The reason that these filters are only available at higher sample rates is that in order for them to be practical they need to operate well outside the audio band.
My choice bounces around a bit, but I often use F6 or F3 with DXD data.
The general concept of the filters is the same throughout the product line, but the actual implementations are slightly different in each product as they are dependent on the capabilities of the underlying processing platform.