Add --dump-fir option to ChiselStage (#3453)

This option will dump the .fir before invoking firtool.

Additional changes:
* Use os.lib for invoking firtool
* Use lazy serialization to avoid holding the entire FIRRTL in
  memory.
* Mix NoStackTrace into FirtoolNotFound
* Fix detection of no firtool

(cherry picked from commit 4db86b2f)

# Conflicts:
#	src/main/scala/circt/stage/CIRCTOptions.scala
#	src/main/scala/circt/stage/ChiselStage.scala
#	src/main/scala/circt/stage/phases/CIRCT.scala