Improve agent code transport

[wouterdb]

Goal

The way code is transported between compile and agent is in need of a rework.

Problems

• Code is now registered per handler type, not per model
  • all handler in the same module get the same code
  • executors now have weird names, as we don't know which module are on there
• Files with the same content but different path go missing on upload

Proposed solution

Build a new code transport mechanism from scratch

In DB structure

1. table for modules
   • a module corresponds to a specific version of an inmanta module (e.g. std)
   • a module has a name, version and a list of requirements
   • we will have two ways of versioning: source based or packaged
     • packaged get a version like 1.3.6 and gets installed per pip (don't have that yet)
     • source based gets a hash (hash of all files)
2. a table of files, joining all files (in the existing file store) to a module (from above)

• file has a path, a hash (link to content) and module/version it is contained in

3. join table joining agent, model version to module, module version

API

• api to request all module/versions for a agent on a specific version
• api to download all files for a module/version
• update codemanager of the scheduler to use them
• api to upload code/version (like current code upload)
• add module/version argument to put_version to populate the join table and to put_partial to check source version remain the same

TODO

• revert part of bugfix for dry-run code loading #8680 (that adds all types for agent)
• remove all tracking of types per agent form the scheduler
• also fixes Make partial export check that the source hashes it has still match the code of its full compile #8089

[wouterdb]

Hey team! Please add your planning poker estimate with Zenhub @arnaudsjs @Hugo-Inmanta @jptrindade @sanderr