This required a bump in the result version: we add a new field to
ResultSuccess.

It doesn't need to know how to write to a file. It store and organizes
data.

I think this is the more common way to do things. It allows you to move
the files around without the symlink breaking.

also fix link path

also lock the directory except when writing to the standard output,
needed for test_generate.py

by mistake from_results was calling latest_bandwidth instead of
earliest_bandwidth

to make it less confussing.
Note that stem call it ed25519_master_key instead of
master_key_ed25519, but dir-spec.txt uses master-key-ed25519, which
is the term chosen here.

also, convert types to int when generating line from str

also complete classes docstring

* Move logic in generate to class
* Add test
* Move 7500 to global constant
* Make read_started_at not fail when conf is None

    * Move V3BWLine to v3bwfile.py
    * Add methods to V3BWLine similar to V3BwHeader
    * Add methods to V3BWLine to implement most of the logic in the class
    * Move functions related to data lines to v3bwfile.py
    * Add functions to generate types of erros
    * Add tests for the class and conftest to read results from file
    * Fix other tests depending on bw lines

so that it close to the header and we can reuse methods/functions

also:
* remove latest_bandwidth outside of class, since it should be
  generated from timestamp
* pass all init arguments as string
* require timestamp argument
* fix tests with changes

they do not need to be in 3rd and 4th posstions.

Also:
* add latest_bandwidth header
* add file_created header
* fix bw files header tests
* disable tests that assume timestamp to be the file creation
time
* if we use ISO 8601 with T as separator,
  datetime format string is not needed

- Create constants for the separators
- Create methods to generate the header text
- Create method to parse the header text
- Fix tests

and include them in the header str.

This reverts commit 628d21a9.

This reverts commit 0b048e91.