How to categorize this PR?

/area auto-scaling control-plane
/kind enhancement

What this PR does / why we need it:
kube-apiservers create a single HTTP2 connection to extension API-servers as gardener-apiserver.
This means that if there are more gardener-apiservers than kube-apiservers in the garden control plane, the "excess" gardener-apiservers do not get any requests.

The consequences of this behavior are:

• Waste of resources since the GAPIs occupy CPU and memory but do not do anything meaningful
• VPA resource calculation is inaccurate. The idle GAPIs reduce the CPU and memory usage so the working GAPIs most likely consume more resources than requested by VPA.

There is a high change that this scenario happens because there is usually more load on GAPIs than on the KAPIs in the gardener-controlplane.

Thus, this PR removes HPA from gardener-apiserver and sets its replicas to 3 in HA case and 2 in the non-HA case. This way we ensure that there are not more GAPIs than KAPIs.
Additionally, the max allowed requests are increased since this is now the only autoscaling mechanism for GAPI.

There could be other options to solve this problem e.g. disabling HTTP2 in GAPIs to force KAPIs to create more connections.
However, it is hard to estimate the consequences of this change and it would need further investigations.

Which issue(s) this PR fixes:
Fixes #

Special notes for your reviewer:

Release note:

`gardener-apiserver` is now deployed with a fixed number of replicas (HA configuration: 3, non-HA configuration: 2). It does not use `HPA` anymore.