Small performance improvement

[s-Nick-s]

Shrinking data.frames in a loop can be an expensive operation.
This PR replaces data <- data[-which(data[[1L]] == R_idx[i]), ] in detect_anoms with a bool index subsetting.
From my tests this makes ad_ts ~20% faster and slightly more memory efficient.

microbenchmark::microbenchmark(
  out1 <- AnomalyDetection::ad_ts(raw_data, max_anoms=0.1, direction='both'), #before
  out2 <- ad_ts(raw_data, max_anoms=0.1, direction='both') #after
, times = 20)

Unit: seconds
                                                                           expr      min       lq     mean   median       uq      max neval
 out1 <- AnomalyDetection::ad_ts(raw_data, max_anoms = 0.1, direction = "both") 3.936969 4.021261 4.214572 4.066235 4.145791 6.737845    20
                   out2 <- ad_ts(raw_data, max_anoms = 0.1, direction = "both") 3.142772 3.208393 3.394106 3.252814 3.330758 4.446146    20