Split up contact_matrix function

[sbfnk]

It's too long and hard to parse - atomised bits should be split off into separate functions.

[sbfnk]

One thing that could/should be done as part of this is to move some of the data (pre-)processing into other functions.

E.g. it would be better along the lines of:

polymod |>
  split_age_into_groups(c(0, 5, 10, 15) |>
  sample_missing("age.group") |>
  contact_matrix()

rather than doing this inside contact_matrix()