[MLIR, CompileDriver] Run entry point signature detection only if needed

Currently compile driver always attempts to detect the entry point signature (see getJITFunction) and prints messages if it can't find one, which is is only relevant for IR string input. Usually we compile Python functions where we already know signatures by design. The suggestions are:

• Change the compiler driver API so we can explicitly ask it to run the inference, and only run it if asked
• Maybe find a better name for this function (JIT used to be the prefix of our entry points, but we changed this)
• Maybe also unify the verbose printing mechanisms, namely emitDIag/emitRemark of MLIR and our own CO_MSG macros. Note, that we might want to keep CO_MSG because emit*s alone get overly verbose when stack traces are enabled. Ref https://github.com/PennyLaneAI/catalyst/pull/317

Start with getJITFunction in mlir/lib/Driver/CompilerDriver.cpp and the CompilerDriver API in mlir/include/Driver/CompilerDriver.h. Determine how callers can explicitly request signature inference and how the diagnostic paths should relate to CO_MSG. Done means inference is conditional for IR string input and unnecessary messages are no longer emitted for Python compilation.