De-duplicate API images that are identical

Our Git repo size is very large. This is mostly from blobs, like our images and videos.

It appears some of the images are identical, such as https://github.com/Qiskit/documentation/blob/main/public/images/api/qiskit/depth.gif. So, it's very inefficient for us to duplicate the same image ~25 times.

An original idea was for historical API docs to use the asset from latest if the blob is bit-for-bit identical. However, there is an edge case there if a new version of latest removes the blob, then all the historical docs are pointing to an asset that no longer exists.

Instead, we could use a folder public/images/api/qiskit/common. If a blob appears in >1 version, we store the blob in /common.

• There is a risk that the same blob filename has multiple versions over time, e.g. version A is in Qiskit 0.19-1.1, then version B is in Qiskit 1.2-1.3+. So, we should probably put something in the file name as a suffix, like the number of bytes or a hash.
• Be careful that the algorithm doesn't slow down gen-api too much. To determine whether an image has a duplicate, we need to inspect every other API version, including the new version we're currently generating
  • Ideally we can do the de-duplication as part of gen-api, rather than a standalone process we sometimes manually run to post-process. With Git repo size, we need to avoid introducing the binary at all because once a blob is saved to Git, it is there forever unless we force push.
  • If we set up a new de-duplication, we need to remember to rewrite the link in the historical API version that now has a common blob

Trace the gen-api entry point and how it writes assets under public/images/api/qiskit across historical API versions. Define a deduplication pass that detects bit-for-bit duplicates, stores shared blobs in common with collision-safe names, rewrites historical links, and avoids adding duplicate binaries before generation completes.