What cause the shape of frequency spectrum convex or concave when I change gain from 34 to 35?
The edges of your spectrum is where the performance of the receiver is hitting the limit. there is a trade-off between flatness and total bandwidth. they could have made the bandwidth smaller to keep it nice and flat but figured, a little extra bandwidth might be nice to have. It's not like this is a spectrum analyzer.
Answer to part 2:
The USRP SDR uses an downconverting IQ modulator before the ADCs. An error in quadrature phase or quadrature levels of the two IQ modulator clock signals or gain and delay balance of the IQ modulator(s) can result in DC artifacts in the spectrum of the IQ ADC data. The ADCs themselves can have a small DC offset in their sampling. Both of those artifacts will show up at 0 Hz in an FFT magnitude spectrum of the IQ data.
Possible answer to part 1:
There is a anti-aliasing low-pass filter in front of an SDRs ADC. If that filters bandwidth is the same as your spectrum plot width, then you might be viewing the transition band of that low pass filter (and any gain stages associated with it) which has non-zero width.
Answer to your first question: As far as I know, the USRP device that you use has a maximum gain of 30 dB. So, I recommend you decreasing the gain a little.
Second one: As you may know, the USRP devices firstly amplify and downconvert the received signal, then sample it. The two first RF stages have a part called "DC choke (maybe RF coupling) " which prevents the DC component of your received signal from flowing through the baseband processing stages (in addition DC signals are not desired in RF circuits, and actually harmful). As you expected, in the ideal case, the peak on the origin should not be seen since the signal is collected with zero mean. However, if your signal contains long non-alternating sequences, RF choke can be penetrated by the DC component and your baseband signal has a DC level as well (DC leakage). Depending on the length of the non-alternating part of your signal, a fading on the DC level may be observed. This is called "DC drop", causing a spectral leakage in the vicinity of origin.