Closes #40055.

After fixing #40017, the datadir files are compressed after 29 days and
deleted after 57. However the total used disk space is less than 3G,
leaving 3G as precaution.

Closes #40044.

After refactoring in #40041, it was forgotten to return the error in
the case a helper was not found, what can happen in test networks.

Closes #40065.

This reverts commit 15da07d6.
Because it's not the default. No additional changes are needed in the
tests.

Closes #40064.

Closes #40056

- reorganize sections
- add diagrams and links
- add pseudocode
- remove math
- correct statements

So that it's more accurate and easier to understand.

into a new file.

and generate .svg from .puml files.
Do not add to the html target since the generated svg images are
not deterministic and will change every time `plantuml` is call.

in different files and link to each other.

Because if the error is not a circuit error, it does not have that
attribute.

and adapt the tests to pass.
\o/

Closes: #40008.

and scripts to run the integration tests with chutney.
It does not replace yet the way integration tests are run.

when connecting to an external tor and they are not already set.

It seems we forgot this option when refactoring in #28738.

to avoid conflict when comparing the options that should be set and the
ones are set, since the SocksPort will be differently in chutney.

to avoid many lines of non useful text the CI.

since they're similar code

After refactoring and making clear when we were using exit(s) that can
exit to all public IPs (and a port) or only some, refactor them
removing the duplicated code and adding the `strict` argument.

When the relay is not an exit, instead of choosing exits that can
exit to all IPs, try with exits that can exit to some IPs, since the
relay will be measured again with a different exit in other loop.

When the relay is an exit, instead of ensuring it can exit all IPs, try
using it as exit if it can exit to some IPs.
If it fails connecting to the Web server, then try a 2nd time using it
as entry to avoid that it will fail in all loops if there is only one
Web server, cause it will be used again as an exit.

Also, the helper exits don't need to be able to exit all IPs. When a
helper exit fails to exit (maybe cause it can not exit to the Web
sever IP), it's not a problem cause in a next loop other exit will be
choosen.

This change of logic also solves the bug where non exits were being
used as exits, because we were trying to measure again a relay that
was used as entry, because it could not exit all IPs, which includes
also the non exits.

Closes: #40041.

At some point all possible errors should be exceptions.

`helper` variable is only used to return error, therefore move it to
the methods that create the path and return the error there.
`our_nick` is not useful for the log, since it is always the same, but
not removing it here.

because `measure_relay` method is too long, confusing and we have had
several bugs in this part of the code.

can exit to some IPs.
To use them in the cases it will be more convenient.

see previous commit

see previous commit

to can_exit_to_port_all_ips, because it's using `strict`, which means
that it allows to exit to all IPs.
It seems more convenient to try first with exits that allow to
exit to some IPs and only try a second time if that fails, because
there are more.

or entry.

Closes #40048

When generating the Bandwidth File as Torflow, use 28 days of past raw
measurements instead of 5, by default.
Also keep the raw measurements for that long before compressing or
deleting them.
And stop checking whether the compression and delete
periods are valid, without checking defaults first and based on
arbitrary values.

Closes: #40017