Migrate Cake.Tool from Autofac to Microsoft.Extensions.DependencyInjection

We're currently not using any advanced features of Autofac (to my knowledge), so we should investigate whether or not we can move to something more lightweight (perhaps the homebrewed DI container we use in `Cake.Frosting`?). I suspect we could reduce the load time for implicit bootstrapping by doing this.

Any investigation of this needs some kind of benchmarking to make sure we're not making it worse (or replace something without an improvement).

Start by tracing Cake.Tool's implicit bootstrapping and comparing its Autofac usage with the homebrewed dependency-injection container in Cake.Frosting. Establish a benchmark for load time before changing the container, then verify that the migration provides an improvement without regressions.

Written by the indexing model from the issue text.