Create events on route updates failures and don't keep node routes to main route table

[MartinWeindel]

What this PR does / why we need it:
Events are now created in the kube-system namespace on route updates, especially if they are failing.
Note, that on normal operation the event is created only once.
If the route update has errors, the event is created everytime. As soon as it recovers an event for normal operation is created one time.
These events will be analysed by a health check of the provider-aws extension.

Additionally, the main table will not contain the routes to the worker nodes anymore. It is sufficient to have them in the route tables for the subnets. If the VPC is used by multiple clusters, keeping the routes in the main table would cause quota issues (RouteLimitExceeded) just sooner. There is a hard limit of 1000 routes per route table according to the AWS documentation, see Amazon VPC quotas

Which issue(s) this PR fixes:
Fixes #

Special notes for your reviewer:

Release note:

Create events on route updates failures and don't keep node routes to main route table

[MartinWeindel]

/invite @ScheererJ @DockToFuture @axel7born

[kon-angelo]

Not sure if it makes sense to make the inverse check of looking at IGW instead of the NAT GWs. Would it make any sense in BYO-VPC cases where there may be other route tables ?

[MartinWeindel]

You are right. Checking of IGW now in main route table.

[ScheererJ]

Is it always required to have an internet gateway? Would it be possible to do it differently in a bring-your-own scenario, e.g. via vpc peering?

[kon-angelo]

Currently this is checked at runtime: https://github.com/gardener/gardener-extension-provider-aws/blob/907425e071477b973e56f319c43786a43ede67af/pkg/controller/infrastructure/configvalidator.go#L119 especially for BYO scenario

[MartinWeindel]

I have simplified the main table recognition by using the name tag.
What do you think? @ScheererJ @kon-angelo

[ScheererJ]

/lgtm

[kon-angelo]

/lgtm