Add a way to pass textual MLIR into the compiler from the Python frontend

[AlexanderViand-Intel]

For debugging, and to support jupyter notebook demos like the ones available via heir_play, the frontend should be able to take in textual MLIR as a string, essentially skipping most of the "Frontend" parts, but still allowing you to use a backend.

E.g., in addition to

@compile()
def foo(x: Secret[I16], y: Secret[I16]):
  return (x + y) * (x - y) + (x * y)


foo.setup() 
enc_x = foo.encrypt_x(7)
enc_y = foo.encrypt_y(8)
result_enc = foo.eval(enc_x, enc_y)
result = foo.decrypt_result(result_enc)

We should also support

ssa ="""
func.func @foo(%x: i16 {secret.secret}, %y: i16 {secret.secret}) -> (i16) {
    %0 = arith.addi %x, %y : i16
    %1 = arith.subi %x, %y : i16
    %2 = arith.muli %0, %1 : i16
    %3 = arith.muli %x, %y : i16
    %4 = arith.addi %2, %3 : i16
    func.return %4 : i16
}"""

foo = compile_ssa(ssa)

foo.setup()
enc_x = foo.encrypt_x(7)
enc_y = foo.encrypt_y(8)
result_enc = foo.eval(enc_x, enc_y)
result = foo.decrypt_result(result_enc)