Feature: improve data set generator

[clarasaja]

This PR updates the generate workflow in HumiTempDatasetGenerator.
It introduces a sub-series concatenation approach and supports multiple occurrences of the same anomaly type within a single series.
The plotting method is also updated to handle these cases.

Changes

New sub-series workflow
Each final time series is built by concatenating n sub-series, each containing zero or one anomaly.

if len(anomalies) == 0:
    n = 1
    
else:
   if auto_repeat_anomalies is False:
      n = min(max_anomalies_per_series, len(anomalies))
      # Then anomalies picked randomly but limited to one per list item
      
   elif auto_repeat_anomalies is True:
      n = max_anomalies_per_series
      # Then anomalies picked randomly with no constraint 

This approach allows a single series to include multiple anomalies, even of the same type.

Support for variable anomaly ratios
Series without anomalies are alternated with series with anomalies according to anomalies_ratio.

auto_repeat_anomalies flag
True: anomalies are automatically reused to fill the requested number per series.
False: anomalies appear only as many times as listed in the anomalies argument. Even with False, you can repeat anomalies by listing them multiple times.

Step anomaly limitation
Limitation for step_uv / step_mv: Short series containing step anomalies are not yet supported.
_divide_time_interval() raises NotImplementedError if time span is too short.

Static plotting method
plot_dataset() is now a static method and uses _plot_func to handle multiple anomalies of the same type.