Use blocks/successors instead of symbols

[thegoldgoat]

For example, regex ab|cd now is compiled to:

module {
  "cicero.flat_split"() {splitTarget = @FLATTEN_0} : () -> ()
  "cicero.match_char"() {targetChar = 99 : i8} : () -> ()
  "cicero.match_char"() {targetChar = 100 : i8} : () -> ()
  "cicero.accept"() {sym_name = "S0"} : () -> ()
  "cicero.match_char"() {sym_name = "FLATTEN_0", targetChar = 97 : i8} : () -> ()
  "cicero.match_char"() {targetChar = 98 : i8} : () -> ()
  "cicero.jump"() {target = @S0} : () -> ()
}

But it would be more efficient (and more idiomatic for MLIR) to use blocks/successors in this way:

cicero.match {
  cicero.flat_split ^bb1, ^FLATTEN_0
^bb1:
  "cicero.match_char"() {targetChar = 99 : i8} : () -> ()
  "cicero.match_char"() {targetChar = 100 : i8} : () -> ()
  cicero.jump ^S0
^FLATTEN_0:
  "cicero.match_char"() {targetChar = 97 : i8} : () -> ()
  "cicero.match_char"() {targetChar = 98 : i8} : () -> ()
  cicero.jump ^S0
^S0:
  cicero.accept
}

with jumps as block terminators, that specify its successor which essentially is the jump target