[Draft] SSA IR Fix Phi Handling

[bsamuels453]

Derived from #1102

This PR changes Phi handling so that Phis are injected at dominance frontiers.

The current implementation adds empty Phi nodes which are later removed and cause variable instances to be missing a definition.

Linked issues:

• Incorrect SSA for indirect call to function modifying global state #434
• Incorrect SSA for conditional after previous dominating conditional #473
• SSA: function parameter use index 1 instead of 0 #1683

These changes also improves the ArbitrarySendErc20 detector, which can now detect test case bad2:

slither/tests/detectors/arbitrary-send-erc20/0.4.25/arbitrary_send_erc20.sol

Lines 44 to 46 in aa3171e

	function bad2(address from, address to, uint256 am) public {
	int_transferFrom(from, to, am);
	}

This PR also fixes a previously unknown issue where the existence of a Phi node is incorrectly interpreted as a variable write when variables_written is constructed. This bug introduced some small false positives, so their unit tests needed to be updated.

[0xalpharush]

These changes also improves the ArbitrarySendErc20 detector, which can now detect test case bad2:

I'm not sure it fixed this as the data dependency tests are failing.

[bsamuels453]

I'm about go to on a project for a month, so if anyone wants to adopt this PR in the meantime, feel free.

Right now it's failing the data dependency tests; specifically the one for tracking data dependencies through function args.

contract PropagateThroughArguments {
    uint var_tainted;
    uint var_not_tainted;
    uint var_dependant;

    function f(uint user_input) public {
        f2(user_input, 4);
        var_dependant = var_tainted;
    }

    function f2(uint x, uint y) internal {
        uint z = x + 1;
        var_tainted = z;
        var_not_tainted = y;
    }
}

generates:

        Function PropagateThroughArguments.f2(uint256,uint256)
                Expression: z = x + 1
                IRs:
                        TMP_1(uint256) = x_0 + 1
                        z_1(uint256) := TMP_1(uint256)

The ssa IR is missing a phi statement for x_0 = phi(user_input_0)

Normally I believe fix_phi would create these phi statements, so I was trying to port that over for the new SSA generation but running into infinite loop issues.