Provide fine grained mechanism for instrumenting a subset of workloads

[ddelnano]

Pixie's instrumentation works at the kernel level since it uses eBPF. This means that all workloads on the machine are instrumented and is generally seen as beneficial since it reduces blind spots. However, there are times where it could be beneficial to scope instrumentation more tightly and to specific workloads. The end user that requested this is interested in lowering the PEM's overhead. In these cases, they want to trace workloads by selecting specific containers, pods or pods in certain namespaces.

Describe the solution you'd like
This end user would like to have runtime toggles that can do the following:

• Specify which containers, pods or pods from specific namespaces should be traced
• Change which protocols are traced
  • Pixie currently allows enabling/disabling protocols via deploy time configuration, but it's not possible to change this at runtime