Remove types, refactor bson packages

[AlekSi]

Problems with the current design:

1. If type conversion is used (for example, from bson.doubleType to float64), validation can be skipped, and we might want to make NaN values invalid in the future.
2. If type conversion is not used, then too much copying might be required.
3. "convert" functions are confusing.
4. Too many types are exported.
5. Remove nolint:deadcode.
6. Reduce memory allocations.

DoD:

• The issue number shouldn't be mentioned in the code anymore (all the places that mention it should be refactored).

[rumyantseva]

I close this issue as the description is not complete/accurate. For validation, we have epic #871, where the complete list of validation tasks is mentioned.

[rumyantseva]

I reopened the issue as its number is mentioned in a few places in the code.

@AlekSi would it be possible to improve the description? Currently, it contains your observations, but it's not straightforward why. E.g. "convert" functions are confusing - what do you mean? What is confusing about them? What needs to be changed?

[AlekSi]

They are horrible. There should be a better solution. I don't know better yet, that's why this task has "not ready" label