drivers: scmi: add use_polling parameter in scmi_send_message

SCMI supports both polling and interrupt modes for message completion.
Previously, the scmi_send_message() API used a 'pre_kernel' flag to
determine which mode to use_polling during the pre-kernel phase and
interrupts post-kernel.

This approach tightly coupled the decision logic with kernel state,
limiting flexibility. In particular, certain power management (PM)
related SCMI APIs require polling mode even in post-kernel context
to avoid unintended CPU wakeups caused by interrupts.

This patch replaces the 'pre_kernel' with a more generic
'use_polling' parameter, allowing callers to explicitly specify
the desired behavior. Typical usage can still rely on k_is_pre_kernel()
to determine polling mode in higher level api, while PM related
calls can directly enforce polling regardless of kernel state.

Signed-off-by: Yongxu Wang <yongxu.wang@nxp.com>