Feature/prevent zombie containers

[abbesBenayache]

🔄 Scénarios gérés :

1. Build long + interruption (Ctrl+C) Le build est interrompu proprement.

2. Build réussi, mais run long + interruption (Ctrl+C)

L'utilisateur peut choisir :
"Yes"
"No"

3. Build réussi, run long, interruption forcée (Ctrl+C x2)

[PierreJeanjacquot]

A few remarks

• we can pass an AbortSignal to dockerode methods, I think it is interesting to abort some methods such as build and push
• we need to give some feedback to the user while running the graceful shutdown procedure
• we need to let the user kill the command (by sending a second SIGINT) when the graceful shutdown is running
• SIGINT handlers must be cleaned as soon as the part of the code requiring a graceful shutdown finished to run
• long build can be tested by adding a RUN sleep x step to the Dockerfile

[PierreJeanjacquot]

see comments

[PierreJeanjacquot]

Registering the listener in the module scope and never cleaning it is problematic because it prevents SIGINT from stopping the process in the whole app.

This creates this bug:

I cannot cancel the command while a long async step is running (in the example TEE transformation step)

We need to register the SIGINT listener just before calling docker and clean it as soon as the docker interaction ends (successfully or with an error).