Parser & internalizing: efficiency fixes.

Parser & internalizing: efficiency fixes.

1) In ParseLazy, we're already using the heap, so we can also tell
AstValueFactory to internalize immediately. This is more efficient.

2) No need to collect the values in values_ and strings_ if they're already
internalized.

3) No need to collect AstValues which are strings in values_ since they don't
need to be internalized (the underlying strings will be internalized
separately).

BUG=429168
LOG=N
R=rossberg@chromium.org

[marja, 2014-11-04 13:48:04]

marja@chromium.org changed reviewers:
+ rossberg@chromium.org

[marja, 2014-11-04 13:48:04]

rossberg@, ptal.

We lost around 7% in MandreelLatency w/ --predictable when I landed the original
"delay internalization CL". This gives us back around 6 % (compared to the
current - lower - score) according to local runs.

[rossberg, 2014-11-04 13:56:44]

lgtm

[marja, 2014-11-04 14:16:07]

Committed patchset #3 (id:40001) manually as 25109 (presubmit successful).