API: allow separating preparation of linting from actual linting

[anderseknert]

While the various WithX methods help set up a configuration for linting, the Lint method is currently where all the processing takes place (parsing, preparing the query, linting, preparing the result).

It would be great if the API offered a way to allow preparing for linting and linting in two separate steps, allowing the more flexible/performant use in some scenarios. Another benefit would be benchmarking, where it would be much easier to differentiate costs of linting if the cost of preparing to lint could be accounted for separately from the cost of linting (i.e. the Rego evaluation).

Obviously, we'll still want the current way to keep working. So I'd imagine the Lint() call would first check if there's a prepare state to start from, and if not, itself call Prepare() (or whatever we want to call it)

linter := NewLinter().
    WithInputPaths([]string{"../../bundle"}).
    WithDisableAll(true).
    WithEnabledRules("opa-fmt").
    Prepare()  // This would be new

rep, err = linter.Lint(context.Background())
if err != nil {
    panic(err)
}