Fix the bucket weight computation

[cshung]

Built on the previous fix, this PR fixes the wrong bucket weight computation. The change is going to be much easier to review once we have done with the previous fix first.

Here are some experiments I ran to show how the weight computation fix works.

This is the base command line: -tc 6 -tagb 80.0 -tlgb 0.2 -sohsi 30 -lohsi 0 -pohsi 0 -sohsr 100-4000 -lohsr 102400-204800 -pohsr 100-204800 -sohpi 50 -lohpi 0 -sohfi 0 -lohfi 0 -pohfi 0 -allocType reference -testKind time

Note that the allocation ratio arguments are not given above, we will vary them in the following cases:

Case 1:
-lohar 10 -pohar 20

Without fix:
sohAllocatedBytes: 17387864196 (20.24%)
lohAllocatedBytes: 22174926546 (25.81%)
pohAllocatedBytes: 46337496222 (53.94%)

With fix:
sohAllocatedBytes: 83407696230 (97.10%)
lohAllocatedBytes: 862837482 (1.00%)
pohAllocatedBytes: 1628816178 (1.90%)

It can be easily seen that the weight was wrong by a large margin, and the fix corrected it.

Case 2:
-lohar 100 -pohar 200

Without fix:
Crashes with System.Exception: Values are not close

With fix:
sohAllocatedBytes: 60546753948 (70.49%)
lohAllocatedBytes: 8375351532 (9.75%)
pohAllocatedBytes: 16977259434 (19.76%)

This repros the Values are not close problem in the base case.

Case 3:
-lohar 200 -pohar 100

Without fix:
Crashes with System.Exception: Values are not close

With fix:
Crashes with System.Exception: Values are not close

The last experiment indicates the bug is not completely fixed, it repros even in the new version, so this is incomplete work, but the original implementation was off by so much that I would like to merge in this partial fix first.

[Maoni0]

this changes the way we survive pretty dramatically, ie, it no longer allows variability in DoSurvive. this issue is to track getting that ability back.