HashTbl avoid backward search in erase()

- The ideas as discussed in email

- The speedup is really there.

Before
loop 0 - Erased 100000 elements:   3.82 s
loop 1 - Erased 100000 elements:   11.45 s
loop 2 - Erased 100000 elements:   19.46 s
loop 3 - Erased 100000 elements:   27.73 s
loop 4 - Erased 100000 elements:   38.74 s
^C

After
loop 0 - Erased 100000 elements (size 2900000 capacity 8388608) 0.01 s
loop 1 - Erased 100000 elements (size 2800000 capacity 8388608) 0 s
loop 2 - Erased 100000 elements (size 2700000 capacity 8388608) 0.01 s
loop 3 - Erased 100000 elements (size 2600000 capacity 8388608) 0 s
loop 4 - Erased 100000 elements (size 2500000 capacity 8388608) 0 s
loop 5 - Erased 100000 elements (size 2400000 capacity 8388608) 0 s
loop 6 - Erased 100000 elements (size 2300000 capacity 8388608) 0 s
loop 7 - Erased 100000 elements (size 2200000 capacity 8388608) 0 s
...