instead of a reference to an Origin entity.

for each entity model `Model`, provide a `models_d` strategy that
produces dicts suitable for using as argument for the `Model.from_dict`
factory method, and reimplement the `models` generator using this
former hypothesis generator.

This is needed to help writing low level tests for model entities.

- add a validator for negative_utc (can be True iff offset is 0),
- update the timestamps_with_timezone hypothesis strategy,
- add low-level tests for it.

This ensures all instanciated model entities have valid types for attributes.

Related to T2308.

(pairing with @vlorentz)

Related to T2310

These are expected to be bytes, not str.

this does not work in the general case since there is no (recursive)
convertion of objects used as model object initialization.

We can only check when using the from_dict() factory.

for better clarity on the code author's intention.

it was still here for bw-compat but should not be necessary any more.

this function does not need to be a local function of the to_dict namespace.

This will allow deduplicating code across loaders.

This generates more realistic contents and avoids spurious HashCollisions when
generating a set of objects using these hypothesis strategies, at the cost of
slightly worse "boundary checking" (i.e. we won't check contents with a length >
4096 bytes).

This lets us generate Content objects directly from a bytestring, with the
proper set of hashes auto-generated from the contents.

Required by loaders, when they can't parse the fullname.

Will be used by loaders.

It just makes more sense that way.

eg. before this change, all leafs would be equal to each other.

They will be used by loaders, so they can deal only with
model objects, instead of having to do the same conversion themselves.

This removes the `data` and `save_path` arguments of `from_file` and
`from_disk`, as data loading is always deferred from now on.
To access it, users are now expected to either open the data files
themselves, or us `.to_model().with_data()`.

Regular files were created, as the 'mode' argument of os.mknod was missing.
However, creating devices requires root; so we can't reasonably do that in
tests.

Instead, we're using /dev/null instead of creating one.
And while we're at it, let's also use /dev/zero (which, if not handled
properly, will result in an infinite read).

It should be cheap enough to do it, and it makes tests easier.

It will be useful to loaders, as they currently load the entire
content in memory before deciding to skip it.

This allows the `swh identify` command to work again.

Close T2288

Can be useful to deduplicate code in swh-storage.

It already is in practice.

The current postgresql model refuses NULL values.

Related to P589

This allows for example to type:

  curl -s https://archive.softwareheritage.org/browse/content/sha1_git:64582b78792cd6c2d67d35da5a11bb80886a6409/raw/ | swh identify
  swh:1:cnt:64582b78792cd6c2d67d35da5a11bb80886a6409	-