The existing FloatFloor/Ceiling functions had an issue with handling
Number.MIN_VALUE (0x1 ==> 5e-324), and would incorrectly return
MIN_VALUE instead of 0/1.  Simplify the entire sequence by using the
available hardware instruction with the appropriate rounding mode
(POS_INF for Ceiling, NEG_INF for Floor).

Add missing FIEBRA/FIDBRA cases to disassembler.

R=jyan@ca.ibm.com,michael_dawson@ca.ibm.com,mbrandy@us.ibm.com
BUG=

Review URL: https://codereview.chromium.org/1853973002

Cr-Commit-Position: refs/heads/master@{#35239}