The TDA change used a binary search (worst case O(log n)) for finding
the right amount of RBs, whereas everywhere else we used a linear search
(O(n)). We generalize the binary search into a function nr_find_nb_rb()
and use it for DL and UL allocation.

Note: Need to maintain same PDSCH and PUSCH TDA list lengths

If the PDSCH and PUSCH TDA lists have different lengths (before this
commit, PDSCH TDA list length was 2, for PUSCH 3), the OAI UE would get
confused: In the special case, the TDA in the DCI for UL was incorrectly
calculated, and a change of the PDSCH TDA list triggered UL never being
decoded correctly.

The TBS calculation is necessary to implement the PF coefficient. To
avoid having to recalculate TDA and DMRS parameters for every UE that we
compare, we pre-calculate (approximate) TB sizes which can be used to
calculate the coefficient.

Until this commit, all user scheduling functionality would search for
the search space and CORESET in every TTI for UL&DL, over and over
again. This commit sets it once in add_new_nr_ue(), and the rest just
uses whatever search space/CORESET is set.

Introduce initialization functions for the FR1 preprocessors: they
can/will be used to pre-compute values needed during UE scheduling.

Add parameters: MCS, Bandwidth (in PRBs), slot scheduling bitmap

To shorten pf_ul() and make it more readable, we put the functionality
of retransmittiong to a UE into a separate function. Thus, it might be
re-used in other scheduling scheme functions (e.g., RR instead of PF).

In order to determine the number of RBs, the pre-processor needs to
calculate the TB size (as it depends on how much of data the UE actually
has to send). Before, the pre-processor would calculate this again,
which obviously is superfluous.

Note: in the DL, we do the same

nr_process_mac_pdu() gets the UE_id instead of the RNTI as parameter to
avoid to search the UE_id after nr_rx_sdu() did that already.

Since nr_process_mac_pdu() is only used/called from one file, remove
from the header file.

Reformat affected code to align more closely with OAI style guidelines.

Due to real-time issue, we might miss HARQ feedbacks. Similarly, I
observed that we get feedback for HARQ processes scheduled for a future
frame (don't know how this can happen).

This commit introduces a function that checks whether the first HARQ
process in the queue corresponds due the current frame/slot combination.
If not, it marks old feedbacks for retransmission.  Furthermore, it also
checks whether the feedback is too early, by not considering this
feedback.