Honor @Order for Map injection sorting order

[quaff]

Fix GH-34688

[quaff]

The failed test is related to this PR, it passes when I run it locally.

java.lang.AssertionError: VerifySubscriber timed out on reactor.core.publisher.MonoFlatMap$FlatMapMain@4c12f54a
	at reactor.test.MessageFormatter.assertionError(MessageFormatter.java:115)
	at reactor.test.DefaultStepVerifierBuilder$DefaultVerifySubscriber.pollTaskEventOrComplete(DefaultStepVerifierBuilder.java:1728)
	at reactor.test.DefaultStepVerifierBuilder$DefaultVerifySubscriber.verify(DefaultStepVerifierBuilder.java:1298)
	at reactor.test.DefaultStepVerifierBuilder$DefaultStepVerifier.verify(DefaultStepVerifierBuilder.java:832)
	at org.springframework.web.reactive.function.MultipartRouterFunctionIntegrationTests.parts(MultipartRouterFunctionIntegrationTests.java:107)

[jhoeller]

There is arguably a semantic problem: Intuitively, a sorted Map would be sorted by key, not by value - whereas in this case, it would be the bean instance values that are to be sorted. That's the main reason we have not implemented this initially but rather left it at bean registration order.

Also, from a modern-day perspective, we mean to discourage Map injection altogether. While it seemed attractive to align it with getBeansOfType in the early days, this is hard to handle in case of actual beans of type Map, and even getBeansOfType is semi-deprecated in favor of getBeanNamesForType plus individual getBean calls these days. It is usually recommendable to use ObjectProvider instead with a clear distinction between bean type and multi-element iteration.

Overall, we intend to refrain from adding further capabilities to Map injection. Thanks for the PR, in any case!