Improve dependencies parsing performance

The dependencies parsing for repos is quite slow.

The reasons for this are:
- To parse dependencies for repo, a repo needs to be cloned into a temporary directory:
- A temporary directory is created before cloning the repo and destroyed immediately after the parsing is finished
- A new process is spawned for each repo to use a tool to parse the dependencies
- Database operations - updating the repo, creating dependencies, repo-dependencies m2m relations.

There are some heuristics in place to prevent unnecessary parsing:
- A revisions hash is stored from the last repo parsing session - so if the revision does not change, we do nothing.

Possible improvements could be:
- Using a process pool to reuse the processes for dependencies parsing sessions.

Start by locating the repository dependency-parsing entry point and the code handling temporary clones, parser process creation, and database updates; no files or tests are named. Compare the current parsing flow with process-pool reuse, but the issue provides no benchmark, test, or explicit completion target for deciding when the performance work is done.

Written by the indexing model from the issue text.