[Bug] Validate topK in Lite group info queries

### Runtime platform environment

Windows, local broker unit tests; no running cluster is required for the regression test.

### RocketMQ version

`develop` at `ff8f6f74c560e391261ccd716707c6d20422e253` (5.5.1).

### JDK Version

Amazon Corretto 8u482; Maven 3.9.11.

### Describe the Bug

`LiteManagerProcessor.getLiteGroupInfo` forwards the request's `topK` directly to the lag calculators when `liteTopic` is absent or empty. `GetLiteGroupInfoRequestHeader.checkFields()` does not validate it.

`LiteConsumerLagCalculator.getLagCountTopK` uses `topK` as the initial capacity of a `PriorityQueue`. Zero and negative values are invalid constructor arguments. Large positive values also control the initial backing-array allocation without a server-side bound.

### Steps to Reproduce

1. Use an existing Lite consumer group bound to a parent topic.
2. Send `GET_LITE_GROUP_INFO` with that group, no specific `liteTopic`, and `topK=0` or `topK=-1`. The equivalent CLI input is `mqadmin getLiteGroupInfo -n -p -g -k 0`.
3. The processor reaches `getLagCountTopK` instead of rejecting the invalid parameter.

The accompanying processor regression test serializes request headers and invokes `processRequest`. It fails on the unmodified implementation because the invalid request reaches the mocked lag calculator. The constructor behavior above is established from the source path; a live-cluster run and large allocations were not attempted.

### What Did You Expect to See?

An `INVALID_PARAMETER` response with a clear valid range, before calling either lag calculator. Requests for a specific LiteTopic should continue to work without `topK`, since that path does not use it.

### What Did You See Instead?

The aggregate query accepts an unchecked heap capacity and reaches the calculator instead of returning a parameter error.

### Additional Context

A small fix can reuse the processor's existing `MAX_RETURN_COUNT` (10,000), accepting `topK` in `[1, 10000]` for aggregate queries. Regression coverage includes null/empty LiteTopic, zero/negative/oversized values, accepted boundaries, and a specific-topic query with the default `topK`.

I searched existing issues and pull requests for `topK` and the Lite lag calculator; the related merged PR #10424 optimizes timestamp lookup and does not add this validation.

Start with LiteManagerProcessor.getLiteGroupInfo and GetLiteGroupInfoRequestHeader.checkFields(), then inspect the existing MAX_RETURN_COUNT and lag-calculator calls. Run the processor regression test that serializes request headers and invokes processRequest. Done means aggregate queries reject topK outside 1–10,000 before either calculator runs, while specific-topic queries still work without topK.

Written by the indexing model from the issue text.