Add some initial performance/scale/capacity tests for impersonation proxy and TokenCredentialRequest API

[mattmoyer]

We're trying to understand roughly how many concurrent users the Concierge can support in the default configuration, as well as what kind of latency overhead is introduced by Pinniped compared to native Kubernetes API operations via kubectl.

We would like some rough numbers we can use to compare performance across releases. We will use initial test results to calibrate our expectations.

Example ideas for tests:

• How many concurrent proxy requests can we make before thing start behaving poorly.
• How much latency is added by using a the Pinniped ExecCredential plugin compared to plain kubectl.

[enj]

Related: we should consider adding protobuf support to our aggregated APIs such as TCR to reduce the network overhead.