The traversal logic for many node types is essentially the same. Use
a single traverse() function for all simple node types instead of
having different ones for each node type. Special traversal code is
only needed for those node types where the traversal logic is
overridden in specific traversers or which do special bookkeeping.

This makes traverser behavior a bit more consistent: InVisit calls are
now done for all node types, including if/else, ternary and loop
nodes. Also false returned from visit function will always skip
traversing the next children of that node.

This reduces shader_translator binary size on Windows by 8 kilobytes.
The added helper functions will also make it easier to implement
alternative more efficient traversers.

Unfortunately this also regresses compiler perf tests by around 2-3%.

BUG=angleproject:2662
TEST=angle_unittests, angle_end2end_tests

Change-Id: I3cb1256297b66e1db4b133b8fb84a24c349a9e29
Reviewed-on: https://chromium-review.googlesource.com/1133009Reviewed-by: Corentin Wallez <cwallez@chromium.org>
Reviewed-by: Jamie Madill <jmadill@chromium.org>
Commit-Queue: Olli Etuaho <oetuaho@nvidia.com>