gh-88037 - Move port validation logic to parsing time

[orsenthil]

Currently urlsplit()/urlparse() accept a netloc with an invalid port without complaint, the error only surfaces later, and only if something actually reads .port:

    >>> p = urllib.parse.urlsplit("http://example.com:oracle/")
    >>> p.netloc          # parses "successfully"
    'example.com:oracle'
    >>> p.path
    '/'
    >>> p.port             # ... raises, but only now, and only if you ask
    ValueError: Port could not be cast to integer value as 'oracle'

This moves that validation to parse time, so urlsplit()/urlparse() raise immediately on a malformed port instead of returning a result that looks valid until .port happens to be accessed.

Supersedes #25774 - Rebase of the old branch wasn't suitable: Lib/urllib/parse.py has seen significant changes since 2021 (WHATWG-compliance work, bracketed IPv6 host validation) that the original patch's diff didn't account for. The original commit is kept in-tact.

---

I was curious on performance of moving this lazy validation to eager, I let the claude run in tiny benchmark.

Benchmark

Measured with timeit (best of 5 runs) on a local ./python build
of this branch (3.16.0a0), comparing the real, unmodified
Lib/urllib/parse.py from main (lazy validation) against this
branch's version (eager, parse-time validation).

_checknetloc in isolation (500k calls each):

netloc	lazy (main)	eager (this PR)	delta
ASCII, no port	54.0 ns	288.4 ns	+234.5 ns
ASCII, with port	53.2 ns	468.2 ns	+415.0 ns
Bracketed IPv6	53.6 ns	575.6 ns	+522.0 ns

Public urlsplit(), which is @functools.lru_cache-wrapped
(128 entries):

scenario	lazy (main)	eager (this PR)	delta
cache hit (same URL reparsed)	124.1 ns	124.2 ns	+0.1 ns (+0.08%)
cache miss (unique URL each call)	2626.8 ns	3131.9 ns	+505.1 ns (+19.2%)

Takeaways:

• When a URL has already been parsed once, the cache absorbs the
  difference entirely — effectively zero added cost.
• For a never-before-seen URL, the added cost is ~500 ns, sub-microsecond
  in absolute terms and negligible next to any I/O the parsed result is
  normally used for (DNS/connect/TLS are milliseconds; this is
  nanoseconds).

• Issue: urllib.parse.urlparse doesn't check port #88037