This way one Tx OOB word may be lost, and tx timestamp won't be generated
causing _wrn_start_xmit: descriptor overflow: tx timestamp pending_ message from
wr-nic driver

- removed declarations of unused signals
- reduced padding_size to bits needed for padding possible runt Ethernet
  frames
- simplified signal assignments and conditions in FSM (e.g. no need to
  check if needs_padding = 1 and padding_size /= 0, because when former
  is true, the latter is always true too)
- moved assignment of outputs validated with other outputs outside FSM
  to simplify the logic (fake rtu decision)

Instead of old fabtic converted to Pipelined WB using ep_rx_wb_master
module from endpoint_private_pkg

- removed ARB_DISABLED state, instead FSM is hold in ARB_START_SCAN
  unless module is enabled
- removed declarations of unused signals (granted_desc_idx, desc_idx_d0,
  check_count, stupid_hack)
- moved assignment of outputs validated with other signals outside FSM
  to simplify the logic (txdesc_current_o/rxdesc_current_o are validated
  by desc_grant_o, and stored in internal registers of nic_rx_fsm and
  nic_tx_fsm)
- rewritten ARB_WRITE_DESC state, so that data output is driven by
  desc_subreg counter instead of separate counter in case statement.
- more exotic regnums in marshalling functions (nic_descriptors_pkg) to
  minimize logic required to output correct data when needed in modified
  ARB_WRITE_DESC state

Just a cleanup to have all subprobrams the same type (functions) in the
package.

[switch-optimization]: wrsw_nic/nic_elastic_buffer: use native almost_full and almost_empty for generating stall

Xilinx FIFOs have native almost_empty and almost_full outputs driven at
configured thresholds. Using them saves few LUTs comparing to stall
generation based on usedw counter as it was done earlier.

It's just a trivial clean up commit, it does not save any resources.

v4-dev: wrsw_nic: sw reset can be a single monostable bit instead of using *_wr signal from the whole register