zzip improvements: constant time deletion, unit tests, better internal apis

[akrieger]

Summary

None

Purpose of change

Some heftier but needed improvements to the zzip implementation.

• deletion requires rewriting the entire zzip to remove one file.
• no unit tests.
• the zzip api was... acceptable. But it could be better.

Describe the solution

• Deleting a file now writes a 'tombstone' which is an empty file entry with a magic constant checksum value 0xDE1337ED (deleeted) which is interpreted as 'this file was deleted'. When reading a zzip from the beginning, if this magic entry is encountered, the program removes any prior instance of the file from internal datastructures. Subsequent entries however are interpreted normally.
• To validate the implementation, since it is new and moderately invasive, I added unit tests for zzip. Specifically, an in-memory 'malloc' based implementation of mmap_file (essentially a mock) which supports all the same operations normal mmap_file supports, but not backed by a real file.
• To make zzip unit testable, I had to refactor the api to remove the concept of path from zzip, because malloc based mmap_file does not have a 'path'. This simplified the internal apis greatly, and also resulted in a better separation of concerns for zzip. zzip does not care about filesystem anymore. Responsibility for filesystem operations like moving files is moved to the callers. Instead zzip largely just operates directly on existing mmap_files. This worked out nicely.

Describe alternatives you've considered

Testing

The new unit tests I added which will run on this PR!

Additional context

[PatrikLundell]

An obvious question is how the magic checksum is guaranteed not to be possible to be generated naturally, or, if it can be, how it is guarded from being considered a "tombstone" (only special case together with empty file, and empty file not generating that checksum?).

[akrieger]

Well, its a combination of magic checksum and zero length file that make up the tombstone. The natural checksum for the empty file is not the magic checksum. yeah what you said.

[github-actions]

Auto-requesting reviews from non-collaborators: @jbytheway

[akrieger]

Found one more bug with the test I was mentally stuck on how to write. Yay!

[akrieger]

Don't merge until the ASAN test runs please.

[akrieger]

Followup #83059