Reland "[turbofan] Weakening of types must weaken ranges inside unions."

Reland "[turbofan] Weakening of types must weaken ranges inside unions."

This relands commit 4c1f4b796d1c455fc6a023abe145a5e48c4b7b1f.

R=rossberg@chromium.org

Committed: https://chromium.googlesource.com/v8/v8/+/2d075e22985f69d775f2a47a77b37e4babb19ffd

[Jarin, 2014-11-13 07:51:49]

Could you take a look, please?

Unfortunately, GetRange is not really monotone, so I removed the test.

The counterexample is Range(1, 2) <= SignedSmall Union Range(1e+10, inf), but
Range(1, 2) <=/= Range(1e+10, inf).

(This also shows that GetRange is really an ugly hack that we will have to
address in a more principled way.)

[rossberg, 2014-11-13 08:05:14]

LGTM

> Unfortunately, GetRange is not really monotone, so I removed the test.
> 
> The counterexample is Range(1, 2) <= SignedSmall Union Range(1e+10, inf), but
> Range(1, 2) <=/= Range(1e+10, inf).

IIUC, this one would also be fixed if we treated number bitsets as proper
ranges. Because then SignedSmall \/ Range(1e+10, inf) would result in
Range(-2^30, inf).

[Jarin, 2014-11-13 09:02:21]

Committed patchset #3 (id:40001) manually as
2d075e22985f69d775f2a47a77b37e4babb19ffd (presubmit successful).