[NPUW]Optimize token rate for dynamic LoRA. (#31742)
### Details:
1. Optimize dynamic LoRA token rate by L0 remote tensor pre-allocation.
2. Solve review comments in
https://github.com/openvinotoolkit/openvino/pull/31433
3. Move some comment functions to utils.
### Tickets:
- *[EISW-179843](https://jira.devtools.intel.com/browse/EISW-179843)*
---------
Signed-off-by: intelgaoxiong <xiong.gao@intel.com>
