Coherent demodulation\detection is usually done via matched filtering in RF signal processing
No, that's not generally true:
- Matched filtering only is an optimal technique if you want to maximize SNR in an AW(G)N channel
- In many, if not most, systems that do matched filtering, that matched filter is realized in baseband, i.e. after coherent demodulation. True matched filters in passband are pretty nontrivial to realize!
Coherent demodulation means you know the frequency and phase of your transmitter exactly at your transceiver.
In radar signals, coherent integration can be done if the phases of the returns are known.
Exactly!
In terms of coherent detection in radar signals, does it make any sense if we still use matched filtering, similarly to RF signals?
Yes, if, and only if, you've got uncorrelated additive noise.
Or, coherent detection means coherent integration in radar signals?
The first is a requirement for the second, I'd say, but not the same.
Notice that matched filtering followed by sampling is essentially a correlation, which is an integral over the pulse shape – so it'd be fair to say that "with a coherent detector, feeding a synchronously sampled matched filter, you build a coherent integrator, giving you the maximum SNR in AWGN channels".