Tedlium not distinct corpora names for the 3 partitions

[Marvin84]

The corpus name for all train, dev, and test partition is the same. This PR suggests a second version of the job that modifies simply the corpus name.

[Atticus1806]

While we are at it: I think we should also touch the part modifying the test/dev set in the Job, such that the new job either does not do this at all or does it via a flag which is disabled by default.
@christophmluscher, since afaik you are/were working on a TED3 Job how did you solve it there?

[christophmluscher]

I solve it very similar to this PR: c.name = f"TED-LIUM-release3-{corpus_key}"

I added a extend segment flag. I removed the updates to the transcriptions completely. Since the segments chagen from rel2 train to rel3 train change and do not occur anymore. for dev/test I removed.

[michelwi]

One could also move much of the duplicated code into a separate function (that only exists in the base class and then the make_corpus function would call this one and set the names accordingly.

We should probably add a deprecation notice to the old class.

Should we also "move" the old class to deprecated?

[Atticus1806]

@michelwi what happens with the apptek hashes here?

[michelwi]

@michelwi what happens with the apptek hashes here?

our recipes depend on #505. If you rebase/merge the current main head the problem should go away.

[Marvin84]

@JackTemaki @michelwi @Atticus1806 @christophmluscher I would like to work on the i6_experiments changes for using the new version, could I get this merged please?

see the comment

[Atticus1806]

LGTM, one thing (which I cant unfortunately manually suggest):
Maybe extend the comment in the old job with the splitting, that this is dangerous / modifies the reference? But approval does not depend on it

[JackTemaki]

As discussed offline, the apostrophe merging is now missing