🐛 remove dead providers from coordinator

[imilchev]

This should minimize the impact of errors we have observed with the scan API. If a provider crashes, it doesn't get properly cleaned up and results in endless loop of failed scans because no provider instance is available. This change makes sure we remove dead providers after each scan, such that the next scan can start with a fresh provider.

This still has the issue that if we have 1 scan with 300 assets and the provider crashes after scanning the first asset, we will get 299 errors for the remaining assets. The issue should be gone with the next scan. A better approach is to detect that a provider instance is dead before connecting to it, if that is the case, we should implement restart logic. That should happen inside the runtime, since we keep a reference to the RunningProvider in all the runtimes for every asset of the scan. We need to make sure we update the same RunningProvider pointer with a new provider instance.

I added a TODO to the code where I think this logic should be implemented.

[github-actions]

Test Results

2 994 tests  +1   2 993 ✅ +1   1m 34s ⏱️ +6s
  329 suites ±0       1 💤 ±0 
   23 files   ±0       0 ❌ ±0 

Results for commit aa17e7d. ± Comparison against base commit 10c73b7.