Atom documents from old deposits usually use these instead of codemeta's

Motivation:

1. It makes it easier to visualize what is actually happening when modifying
   the graph, by working explicitly on triples instead of a JSON-LD (a tree
   serialization of the graph).

2. Remove the need for the hacky `merge_values()` function (and possibly
   `merge_documents()` in a future commit)

3. It also catches malformed data exactly where it is added in the document
   (the call to rdflib.Graph.add()) instead of at the end of the mapping
   when running compaction/expansion.

Downsides:

1. Tests are clunkier, because they relied on deterministic order of
   unordered lists; but rdflib does not guarantee it

2. Code is longer

3. Extra dependency (which we will need at some point if we want to
   import from RDF datasets, anyway)

Their order is nondeterministic, it just happens to work with
the way we use PyLD.

This parameter was only used to execute extra code before
`normalize_translation` is called. This caused some duplication, and
will not work when switching to a non-JSON-LD internal representation.

Removing it also makes the code of mappings more consistent, by removing
specific field handling from their implementation of the `translate`
method itself.

They cause postgresql to crash because it does not allow them in text fields.

They are seemingly only present accidentally in source documents;
so stripping them does not really impact the quality of metadata.

When capture_exceptions=False, the indexer's caller reports the exception
to Sentry itself.

However, because tags were added by indexers within a scope internal
to the indexers; the scope was closed before returning to the caller,
so these tags were actually not sent to Sentry.

Without these tags, it is often impossible to find what object
caused a given crash without guesswork based on the object's
content and swh-graph.

Invalid URLs are a common source of crashes

Detected through T4406

Related to T4412

So future services to be deployed can match on that name.

Related to T4406

The indexer language has been no longer running in production for years (and the related
indexer code has been pruned years ago as well). Those related endpoints are not
consumed by anyone. So we can drop those.

That will ease code maintenance and make the ci gain some time when running the overall
tests.

Related to T4273

It's been no longer running in production for years. That will ease code maintenance.

Related to T4273

Related to T4273

Related to T4273

intrinsic mappings use the short name, instead of the class name.

1. indexers call themselves directly instead of going through the scheduler
2. metadata is attached to directories instead of revisions

It makes resulting documents (usually) shorter, and tests more readable.

detect_metadata_files and extrinsic_metadata_formats (respectively) are somewhat
mutually exclusive, so it does not make much sense to have them in the same
class and MAPPINGS dict

It is already set by _translate_dict itself.