fix: detect query string / fragment directly after a domain

[abhu85]

Problem

Fixes #516.

A query string or fragment that comes directly after the domain, with no path slash, is dropped:

import { find } from 'linkifyjs';

find('example.com/?foo=bar')[0].href; // http://example.com/?foo=bar  ✅
find('example.com?foo=bar')[0].href;  // http://example.com           ❌ (query dropped)
find('example.com#frag')[0].href;     // http://example.com           ❌ (fragment dropped)

Cause

In the parser state machine, the DomainDotTld (e.g. example.com) and DomainDotTldColonPort (e.g. example.com:8080) states only transition into the query-accepting Url state on tk.SLASH:

tt(DomainDotTld, tk.SLASH, Url);
tt(DomainDotTldColonPort, tk.SLASH, Url);

There's no transition on ? (QUERY) or # (POUND), so a ?/# without a preceding / ends the match at the bare domain. The scheme-prefixed path already handles this (tt(SchemeColon, tk.QUERY, Url)), which is why http://example.com?x=1 works but example.com?x=1 doesn't.

Fix

Add the missing transitions, mirroring how the Url state already treats these tokens:

tt(DomainDotTld, tk.QUERY, UrlNonaccept);  // `?` is non-accepting -> trailing `?` stays trimmed
tt(DomainDotTld, tk.POUND, Url);           // `#` is accepting
tt(DomainDotTldColonPort, tk.QUERY, UrlNonaccept);
tt(DomainDotTldColonPort, tk.POUND, Url);

Because QUERY routes to UrlNonaccept (consistent with qsNonAccepting), sentence punctuation like have you seen example.com? is not over-matched — the trailing ? is still trimmed.

Test plan

• Added parser cases: example.com?foo=bar, ?a=1&b=2, :8080?x=1, #section are single Url tokens; have you seen example.com? keeps the ? as trailing text.
• Updated one existing case (google.com?q=asda) whose expectation encoded the old drop-the-query behavior — it's now a single Url, same token types.
• New cases fail on main (5 failing) and pass with the fix.
• Full suite 394 passing; eslint clean.