Average $f_tile$ computation in randoms vs data

$\left(N_{tile})$ (`fttl`) is a factor in the density $n(z, N_{tile})$ that goes into the FKP weights (Section 7.3 of [Ross et al 2025](https://arxiv.org/abs/2405.16593)). In [common_tools.py#L992](https://github.com/desihub/LSS/blob/c797b5075b94d0ac1a3c55c35f30185c1d0df773/py/LSS/common_tools.py#L992), it is computed as an average in randoms, but in [mkCat_subsamp.py#L351](https://github.com/desihub/LSS/blob/c797b5075b94d0ac1a3c55c35f30185c1d0df773/scripts/mkCat_subsamp.py#L351), it is computed as an average in data. I know that $f_{tile}$ (`FRAC_TLOBS_TILES`) is computed in data for each set of overlapping tiles (`TILES`) and then propagated to randoms. Still, doing its average on randoms should provide a more accurate area (geometric) average, whereas I am not sure if the average on the data clustering catalog has a clear-cut interpretation. But hopefully, the difference is very slight.

This probably should not be changed until DR2 is finished.