Share the preamble in switch interpreter.

Move the preamble code to the main loop to avoid replicating it.

This halves per-file compile time to 0.5 min.

It saves 225k(arm32) and 348k(arm64) from libart.so.

Performance difference is within noise (golem using the switch).

Test: "test.py -b -r --interpreter --host" with switch interpreter.
Change-Id: I2615b3568eb79560d2d342ac3dd28f6269ee5e74