Adapt the memory merging pass from the legacy pipeline into venom
pipeline with additional improvements. It merges instructions that copy
data or zero memory into a single copy instruction if possible (with the
exception that if the copy is exactly 32 bytes, it uses a load/mstore
pair to save 1 byte / 3 gas). It also handles sequences of interleaved
and out-of-order copies, even though these are not currently generated
by the frontend.

Examples:

input1:
```
    %1 = mload 0
    mstore 1000, %1
    %2 = mload 32
    mstore 1032, %2
```
output1:
```
    mcopy 1000, 0, 64
```
input2:
```
    mstore 100, 0
    mstore 132, 0
```
output2:
```
    %1 = calldatasize calldatacopy 100, %1, 64
```

It handles copies from memory (mload/mstore/mcopy) and calldata
(calldataload/calldatacopy). For zeroing memory it uses `calldatacopy`
from the `calldatasize` offset.