[VL] LowCopyNettyJniByteInputStream does not work in uniffle netty mode

### Description

Shuffle reader does not use `LowCopyNettyJniByteInputStream` as expected in uniffle netty mode, but still uses `OnHeapJniByteInputStream`.

uniffle client conf:
```
spark.rss.client.type=GRPC_NETTY
```

### Gluten version

None

Start by tracing the shuffle reader's input-stream selection when spark.rss.client.type=GRPC_NETTY, comparing where LowCopyNettyJniByteInputStream and OnHeapJniByteInputStream are chosen. Done means uniffle netty mode selects the low-copy stream as described and the behavior is covered by an appropriate regression check.

Written by the indexing model from the issue text.