Fix hostname blocking

[na--]

This should close #1721 (cc @curvedriver). There were 2 separate bugs... 😞 Wildcards didn't catch 0-length matches, and when you blocked both a domain and a sub.domain, the HostnameTrie stopped blocking the domain because the len(t.children) == 0 check was no longer true (wrong assumption that you can't block both a top-level and a sub-domain). This script more succinctly demonstrates the issue, running it with k6 0.29.0 will fail all checks, whereas with this PR they all pass:

import http from 'k6/http';
import { check, group } from "k6";

export const options = {
    blockHostnames: ['*httpbin.org', 'test.k6.io', 'k6.io']
}

export default function getFromBlockedHostname() {
    [
        'https://httpbin.org/get?my_param=my_value',
        'https://k6.io/',
    ].forEach(url =>
        group(url, () => {
            check(http.get(url, { redirects: 0 }), {
                "status is 0": (r) => r.status === 0,
                "error_code is 1111": (r) => r.error_code === 1111,
            });
        }));
}

[mstoykov]

Shouldn't this become ... false if it goes beyond this ?

[na--]

I don't understand - the next line is return nil, so it can't go beyond this 😕

[mstoykov]

I meant that if on sequential call to insert on a node that is already a leaf ... we won't unleaf it ... which I now understand is exactly what this is fixing (but I needed to read the whole code ...).

I am somewhat of the opinion that if there is "*valid.test" and "test.valid,test" we should probably not add the second one to tree ... and/or give out a warning/error that there is a more general glob already.

But this can be done as an optimization in later PR and I am not particularly opinioanted about it ...

[imiric]

LGTM, nice work! 👍