Allow custom StructuredOutputConverter(s) to participate in Native Structured Output

[filiphr]

This builds a bit on top of #5412.

The purpose of this is the fact that we extensively use the structured output with our own StructuredOutputConverter(s), which do not extend from BeanOutputConverter. Unfortunately, we now need to do something like

if (StringUtils.isNotEmpty(outputSchema)) {
    clientRequestBuilder.context(ChatClientAttributes.STRUCTURED_OUTPUT_SCHEMA.getKey(), outputSchema);
    clientRequestBuilder.context(ChatClientAttributes.STRUCTURED_OUTPUT_NATIVE.getKey(), true);
}

instead of just passing our output converter to the ChatClient. I came up with the workaround because I read the code and how everything works under the hood in Spring AI.

From what I could see the BeanOutputConverter is not really needed for this to work, since any StructuredOutputConverter should be able to provide a JSON Schema if they want to.

Would appreciated your feedback on this.

[filiphr]

Adding some explanation for some changes that are not entirely related to the initial request. In theory this could lead to some problems outside of the proposed changes here since the contract between the context and the underlying implementations differs.

If you prefer I can split this PR into 2 to handle the contract difference.

[filiphr]

This is needed due to the fact that Map.copyOf does not allow null values, whereas the contract of the context is Map<String, @Nullable Object>.

[filiphr]

Same reason as the change of Map.copyOf

[filiphr]

Same reason as the Map.copyOf. ConcurrentHashMap does not allow null values.