Add kernel compilation pass for ORC JIT remote execution

[rniczh]

Context:

Description of the Change:

ENABLE_REMOTE=ON make all

Take x86_64 as an example:

import os

import pennylane as qp

import jax.numpy as jnp
from catalyst import qjit
from catalyst.pipelines import default_pipeline

REMOTE_TARGET_TRIPLE = "x86_64-linux-gnu"
REMOTE_ADDRESS = "ADDRESS:PORT"
WORKSPACE = os.path.dirname(os.path.abspath(__file__))


def _pipeline_with_cross_compile():
    stages = default_pipeline()
    cross_compile = (
        "cross-compile-remote-kernels"
        f"{{workspace={WORKSPACE} "
        f"target={REMOTE_TARGET_TRIPLE} address={REMOTE_ADDRESS}}}"
    )
    for name, passes in stages:
        if name == "MLIRToLLVMDialectConversion":
            passes.insert(0, cross_compile)
            break
    return stages

def main():
    dev = qp.device("lightning.qubit", wires=2)

    @qjit(pipelines=_pipeline_with_cross_compile(), keep_intermediate="pass")
    @qp.qnode(dev)
    def circuit(theta, weights):
        qp.RX(theta, wires=0)
        qp.RY(weights[0, 0], wires=0)
        qp.RZ(weights[1, 1], wires=1)
        return qp.expval(qp.PauliZ(0)), qp.probs(wires=[0, 1])

    print(circuit(0.5, jnp.array([[0.1, 0.2], [0.3, 0.4]])))


if __name__ == "__main__":
    main()

TODO:

• remove remote_close, it's useless.

Benefits:

Possible Drawbacks:

Related GitHub Issues:
[sc-119223]

[rniczh]

We split this PR to the following PRs for reviewing.
#2909
#2910
#2915
#2916
#2889
#2904