all_true, any_true, bitmask, and extract_lane all replace a simd type
with single scalar, so we don't need to set nullptr for the extraneous
lane, and can set num_lanes to 1.

So for a call node with any of the above nodes as one of its value, we
were incorrectly replacing the extract lane node with 4 nodes - it
should only be 1 scalar node.

Drive by fix for LowerShift, it needs to check if the shift value node
has replacement (e.g. if it is an extract lane node).

Also we weren't setting the lowered type of a kPhi node. This can be a
problem if the output of a Phi node is used by a F32x4ExtractLane and
I32x4ExtractLane, which is possible (and type correct since both take
v128). This will require that the output be both a float and a int,
which is impossible. So fix it to Int32x4, and the uses of the phi nodes
will convert appropriately.

Added a test for the call node. Also tested this code by running one of
our benchmarks, which exercises the Phi logic.

Bug: v8:10507
Change-Id: I01a76b349ba9d0e157b64e737bdb4c70f96aa954
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2536952
Commit-Queue: Zhi An Ng <zhin@chromium.org>
Reviewed-by: Bill Budge <bbudge@chromium.org>
Cr-Commit-Position: refs/heads/master@{#71218}