Several problems are present.

The first is that _write returns 0 instead of the
number of samples. We solve it by returning
nsamps.

The second is that _read may return less samples at
the beginning and we don't want to exit for that.
We solve it also by returning nsamps.
(We still need to log more in this, to be done in the
next commit.)

The third is that after initialization we don't send
anything for a while, time for the softmodem to finish
its init. This generates lots of "RX overrun".
We solve it by disabling TX and RX modules after init
and then in trx_brf_start we activate them again (and
also call bladerf_sync_config, which seems to be
mandatory, and bladerf_calibrate_dc, which may be avoided,
perhaps).

Maybe not the end of the story. Sometimes it works, UE connects,
traffic is fine (tested only with 5MHz). Sometimes it does not,
UE does not connect, or it connects but then traffic is bad,
especially uplink.

To be refined.

those popped up thanks to previous commit

No need for this special case. Plus, it fails with the work
from Nokia on compilation cleanup.

We have a strange problem with the timestamp of the very
first packets received from the LimeSDR that leads to
a complete failure of the eNB.

A simple workaround is to discard the first RX packet
just after a "start" of the device.

We also change the value of "tx_sample_advance" to a more
reasonable value. The problem of timestamp was "solved"
differently in the past and at the time we needed to put
a much higher value.

This may have to be refined at some point.

Bug reported by Bruno Mongazon from Nokia.

This reverts commit ce14050b, reversing
changes made to 0ada1e9e.

added prach buffer for if4p5 processing in RRU. This fixed a critical bug which was seen for bandwidths greater than 5 MHz.  RRU tested with 5/10 MHz on 1.5 GHz Atom.