validation changes

[LoannPeurey]

Is your feature request related to a problem? Please describe.
change validation and metadata behaviour

Describe the solution you'd like

• remove dialect from language column (maybe add a dialect column instead)
• start_time NA does not produce warnings
• unknown sex entries should give '?' in overview for number of children, and not fail on non string data
• format custom should be accepted in annotations
• json (overview) serialization should have an encoder for sets (from metannots parameters) stack ovf
• None (empty values) for some fields (monoling, child_sex etc) happend when we don't know, maybe remove warning or make it clear that this is a sign of lack of knowledge

[LoannPeurey]

Then: optimize validation as well as init of objects:

• redefine times where files are read (.read()) for projects and annotation manager
• optimize init of objects, let go of too many checks
• optimize validation of dataframes:
  • Less checks
  • better validation methods => (use pandas vectorized checks and or library like pydantic => use pandantic validation with annotated types using pydantic fields
  • more choice in the granularity of the validation