[intel npu] General NPU optimizations for set_tensor API

[MirceaDan99]

Details:

• Add optimization for logger:
  • Avoid repeated creation/deletion of Logger objects in each intel_npu::Config::get() intel_npu::OptionDesc::get() methods by adding a Logger member in each of the Config and OptionDesc classes
  • Logger class was refactored to accept only const char* variables for its internal name, ctors/methods using std::string_view and std::string params were explicitly deleted. By normal usage, Logger objects get their name statically initialized e.g. _logger("ZeroEngineBackend", ... and there would be no need for a runtime name container
• Optimize std::string to std::string_view:
  • Statically initialization of NPU options is also an optimization opportunity, additional std::string_view->std::string copies were found at Config::get(): const auto it = _impl.find(Opt::key());. Improved my changing internal Config implementation map's key to std::string_view
  • There still would be other std::string->std::string_view covered in "Aggressive" optimization of std::string to std::string_view in configs/properties yet to be discussed as they might be out of scope for the purpose of this PR
• Optimize intel_npu::SyncInferRequest::find_port and ZeroInferRequest::set_tensor methods + Small optimization of IODescriptor objs copies from ZeroInferRequest::infer_async and ZeroInferRequest::get_result:
  • General unintended copies of functions that return references

Tickets:

• E197758