Fix request: Sparse sampling with NA values

[Gero1999]

When running sparse NCA, pk.nca() aborts if the concentration data contains any NA. A single NA anywhere is fatal; it's not about the pooled mean being undefined.

library(PKNCA)
d_conc <- data.frame(
  subject = c(1, 2, 3, 4, 5, 6),
  time    = c(0, 0, 4, 4, 24, 24),
  conc    = c(0, 0, 5, 7, 3, NA)
)
d_dose <- data.frame(subject = 1:6, time = 0, dose = 100)
o_conc <- PKNCAconc(d_conc, conc ~ time | subject, sparse = TRUE)
o_dose <- PKNCAdose(d_dose, dose ~ time | subject)
o_data <- PKNCAdata(o_conc, o_dose,
  intervals = data.frame(start = 0, end = 24, sparse_auclast = TRUE))

pk.nca(o_data)   # Error: ... Assertion on 'conc' failed: Contains missing values (element 6).

What I'd suggest: handle NAs when forming the sparse mean rather than asserting them away. Options, roughly in order of preference:

• In as_sparse_pk(), allow missing concentrations (any_missing_conc = TRUE) and drop NA records when building the per-timepoint groups, so each mean is computed over the available subjects.
• In sparse_mean(), the per-timepoint mean currently uses mean(current_time$conc) with no na.rm — using na.rm = TRUE (and adjusting the sparse n/df for that timepoint accordingly) would make it robust.
• If a timepoint has all subjects missing, drop that timepoint from the mean profile rather than erroring.
• At minimum, if you'd rather keep the strict assertion, a clearer error message (which subject/timepoint, and guidance to remove NAs for sparse) would help a lot.

For now in aNCA we're working around it by dropping NA-concentration rows before the sparse calculation (unless the user imputed NAs), but it'd be great to have this handled upstream by PKNCA. I will open a small PR proposal for this