[Platform] Make DeferredResult and RawHttpResult safe to dump()/dd()

[alireza-aminzadeh]

Q	A
Bug fix?	yes
New feature?	no
Docs?	no
Issues	Contributes to #2280
License	MIT

Every HTTP-based ModelClient wraps its injected client with
EventSourceHttpClient (needed to decode SSE streaming responses), so the
response held by RawHttpResult is always a Symfony HttpClient
AsyncResponse, itself wrapping a TraceableResponse whenever HTTP client
profiling is enabled, which is the default in a full-stack app's dev
environment. Neither class has a dedicated VarDumper caster, and both keep
their network stream alive until it is explicitly consumed.

dump()'ing or dd()'ing a DeferredResult right after
PlatformInterface::invoke(), before the result is ever read, therefore
reflects into that live, not-yet-consumed response instead of a safe
snapshot: closures bound to the response, its HTTP client and its internal
chunk-accounting state all get expanded, and poking at that state from
outside the client's own bookkeeping is exactly what the
AsyncResponse/TraceableResponse pair is not designed to tolerate.

This PR adds __debugInfo() to RawHttpResult so it never exposes the live
response in a dump, replacing it with a short, static summary of its type
instead of letting the dumper walk into its internals. It also adds
__debugInfo() to DeferredResult so a dump surfaces the conversion state
(pending, converted or failed) at a glance, without ever calling
getResult() itself, so inspecting a pending result cannot trigger a
conversion as a side effect.

Tests assert that __debugInfo() never calls any method on the wrapped
response, that the httpStream property stays visible, and that
DeferredResult's reported state matches pending/converted/failed without
triggering a conversion.

No CHANGELOG.md/UPGRADE.md changes, since this is a bug fix only.

[carsonbot]

Hey!

To help keep things organized, we don't allow "Draft" pull requests. Could you please click the "ready for review" button or close this PR and open a new one when you are done?

Note that a pull request does not have to be "perfect" or "ready for merge" when you first open it. We just want it to be ready for a first review.

Cheers!

Carsonbot

[chr-hertel]

Nice idea, thanks @alireza-aminzadeh! Haven't tested it yet, but I think for the DeferredResult it would be nice to keep more of the instance's state, e.g. error message, metadata, options

[alireza-aminzadeh]

Thanks for the feedback! Pushed a commit that extends DeferredResult::__debugInfo() with the three pieces of state you mentioned, none of which touch $rawResult any more than the existing state entry does:

• options — the array the DeferredResult was constructed with (a plain, readonly, always-initialized constructor argument, e.g. the model options a call site passed to invoke()).
• metadata — $this->getMetadata()->all(); empty until conversion succeeds, and even then only ever holds what a ResultConverter or a stream listener explicitly added to it (e.g. token_usage), never the raw response itself.
• error — $conversionFailure?->getMessage(), null unless the state is failed; reading the message doesn't walk the exception's stack trace.

Also widened the @return docblock to an explicit array shape and added one test per new entry (testDebugInfoIncludesOptionsWithoutConverting, testDebugInfoReportsEmptyMetadataAndNullErrorWhilePending, testDebugInfoIncludesMetadataAfterGetResult, testDebugInfoIncludesErrorMessageAfterConversionFailure), covering the pending/converted/failed states.

Rebased on the latest main while at it. The full platform suite (908 tests) and PHPStan pass locally. Ready for another look whenever you have time!

[chr-hertel]

just realizing that the API design here is not that great because the getMetadata() has the side-effect of creating metadata if not existing. usually not an issue, but maybe misleading while debugging.

how about:

Suggested change

	'metadata' => $this->getMetadata()->all(),
	'metadata' => $this->metadata?->all(),

keeps the state more honest

[chr-hertel]

Will patch that minor finding while merging - thanks @alireza-aminzadeh!