replayer: Filter out colliding contents when replaying

Now that we have the colliding contents details [1], we can be less lossy when replaying contents to another storage.

Prior to this commit, we would not write the batch of contents to the destination if a collision was detected [2]. Now we keep those out of the loop if any. Effectively, not writing the colliding hashes to the destination storage.

We still have to decide what to do for the colliding contents... For that, we still have the logs referencing the colliding contents though (and sentry [1]).

Note that could be a shared behavior, in a retrying proxy for example?

• [1] https://sentry.softwareheritage.org/share/issue/db05509e5a9d40dc8fe782be90f2f712/

• [2] Related to !114 (closed)

Test Plan

tox

---

Migrated from D2800 (view on Phabricator)