Introduce a shared memory cache for PrewriteBuffer between different replica

### Search before asking

- [x] I searched in the [issues](https://github.com/apache/fluss/issues) and found nothing similar.

### Motivation

When testing the stability of `PrimaryKey tables`, a frequent issue arises where exceptionally high write traffic peaks cause delayed log synchronization. This leads to continuous accumulation in the KV's `PrewriteBuffer`. Since the current `PrewriteBuffer` lacks a back-pressure mechanism, has no memory cap, and utilizes JVM heap memory, it can easily trigger OOM errors, resulting in `TabletServer` crashes.

Therefore, Fluss needs to implement a back-pressure mechanism to prevent this scenario. A promising approach under consideration is to allocate a fixed-size block of off-heap memory shared across different replicas, while also imposing per-replica memory limits.

### Solution

_No response_

### Anything else?

_No response_

### Willingness to contribute

- [x] I'm willing to submit a PR!