stage2: correct comptime-known inferred alloc type, plus comptime array init sentinels

[mitchellh]

Two changes in here, both needed to get the tests passing in this PR.

At least 3 more passing tests!

1. For inferred allocs with values known at comptime but not forced comptime (i.e. not in a comptime {} block), we were setting the incorrect type, causing cascading errors. We can now set the type to the bitcast type. The previous comment noted we can't do that due to constness, but @Vexu's recent make_const_ptr work fixes this.

2. A mistake from my prior PR on sentinel array initialization, I was not adding the sentinel to the end of the array value for comptime. Bonus: this solidifies that adding the sentinel as the final op in the array_init_sent ZIR is correct.

Tests used during debugging/development are below. First two are the inferred alloc commit, last one is the sentinel bug. The real behavior tests exercise both at once.

test "inferred alloc with sentinel" {
    const ary = [_:0]u8{42};
    const ptr: [*:0]const u8 = &ary;
    try expect(ptr[1] == 0);
}

test "inferred alloc with sentinel at comptime" {
    comptime {
        const ary = [_:0]u8{42};
        const ptr: [*:0]const u8 = &ary;
        try expect(ptr[1] == 0);
    }
}

test "comptime sentinel array init" {
    comptime {
        var ptr: [*:0]const u8 = &[_:0]u8{1};
        try expect(ptr[1] == 0);
    }
}

[andrewrk]

Nice work!

Tests used during debugging/development are below. First two are the inferred alloc commit, last one is the sentinel bug. The real behavior tests exercise both at once.

For future reference, I would consider it an improvement if you took any behavior tests that test too much at once and meaningfully split them into multiple independent tests like this.