rpk: don't require --partition when using %p in topic produce

[telegraphchi]

rpk topic produce only enabled franz-go's manual partitioner when
--partition was set to a non-negative value, so a %p directive in the
--format string was silently ignored unless -p was also passed. As the
issue notes, the flag's presence — not its value — was the real switch,
which is confusing:

$ rpk topic produce t -f '%p %v\n'        # %p ignored, default partitioning
$ rpk topic produce t -f '%p %v\n' -p 1   # %p honored

This PR enables the manual partitioner whenever the input format references the
partition via %p, so %p works on its own. A small helper detects a real
%p directive while respecting the %%, %{ and %} escapes (so %%p is
literal text, not a directive). The help text and the --partition flag
description are updated to match.

Fixes #17697

Backports Required

• none - not a bug fix

Release Notes

Improvements

• rpk topic produce now honors a %p partition directive in --format
  without requiring the --partition flag.

[WillemKauf]

Docs here seem to indicate this is expected behavior. They will at the very least, need updating, should we alter the behavior like this

[telegraphchi]

Good catch — agreed the docs need updating. That examples page describes today's behavior (where %p requires a non-negative --partition), and this PR does change it, so I've already updated the in-repo --help/long text and I'm happy to follow up with a PR to redpanda-data/docs for that page.

On whether we should change it: the motivation (#17697) is that the current behavior is surprising — when %p is present the --partition value is unused; only the flag's presence decides whether %p is honored, so %p without -p silently no-ops. Since the only reason to put %p in a produce format is to direct records by the parsed partition, enabling the manual partitioner whenever %p appears makes it behave the way the format implies. It doesn't affect any invocation that already passed --partition (manual partitioning was already on there) — it only fixes the %p-without---partition case that previously did nothing.

Happy to gate it differently if you'd prefer (e.g. keep it flag-driven but document clearly), but this seemed like the least-surprising option. Let me know your preference on the docs follow-up too.

[WillemKauf]

Good catch — agreed the docs need updating. That examples page describes today's behavior (where %p requires a non-negative --partition), and this PR does change it, so I've already updated the in-repo --help/long text and I'm happy to follow up with a PR to redpanda-data/docs for that page.

I don't mind if your code is AI generated but I wish your responses to me weren't, just FYI.

I think the new behavior is probably better, but @r-vasquez is probably the person to make that call.

[telegraphchi]

Thanks Willem, appreciate you taking a look. Agreed it's @r-vasquez's call — I'm happy either way. Whichever way it lands, I'll update that docs page to match.