[v0.11.0] [P/D] force with_prefill true after allreduce in kv producer

[liziyu179]

What this PR does / why we need it?

force with_prefill true after allreduce in kv producer

Does this PR introduce any user-facing change?

How was this patch tested?

[github-actions]

👋 Hi! Thank you for contributing to the vLLM Ascend project. The following points will speed up your PR merge:‌‌

• A PR should do only one thing, smaller PRs enable faster reviews.
• Every PR should include unit tests and end-to-end tests ‌to ensure it works and is not broken by other future PRs.
• Write the commit message by fulfilling the PR description to help reviewer and future developers understand.

If CI fails, you can run linting and testing checks locally according Contributing and Testing.

[gemini-code-assist]

Code Review

This pull request aims to fix issues related to prefill for the KV producer. The changes include moving a code block in _dummy_run to ensure with_prefill is set correctly before being used, and adding a check in save_kv_layer to only proceed if there are requests. The code movement in vllm_ascend/worker/model_runner_v1.py correctly fixes a bug. However, the change in vllm_ascend/distributed/mooncake_layerwise_connector.py introduces a critical AttributeError due to a typo. I've provided a comment with a suggested fix for this issue.

[gemini-code-assist]

There's a typo in the attribute name. The MooncakeLayerwiseConnectorMetadata object has an attribute requests (plural), not request. This will raise an AttributeError at runtime.

Additionally, to check if the dictionary is not empty, you can use the dictionary directly in a boolean context, which is more Pythonic.

Suggested change

	if self.kv_role == 'kv_producer' and connector_metadata.request.keys():
	if self.kv_role == 'kv_producer' and connector_metadata.requests:

[MengqingCao]

This is a backport of #3768 and #3849