Have sign bit indicating error in error union

[ArborealAnole]

const eu = error{ problem1, problem2, problem3 }!u63;
const out = std.io.getStdOut().writer();
try out.print("error union size: {}\n", .{@sizeOf(eu)});
try out.print("test size: {}\n", .{@sizeOf(?u63)});

output:

error union size: 16
?u63 size: 16

Why not just store the error codes as two's complement negative integers in the same space the u63 would be stored in the error union? That way the error union size would be only 8 bytes.

In the error union, it would also be possible to try fitting in error associated-values (which were proposed in #2647), though that could involve more complexity to varying degrees. E.g. you could truncate the negative i64 error code to i32 and have room for a u32 error-associated value in this case.

Also, I know I might eventually be able to use a proposed arbitrary-range integer type as in #3806 with 0 as sentinel for absence of value, but if for some reason I wanted 0 to be a valid value while sacrificing 2^63 valid values, the absence of a value for the optional u63 could be indicated by the sign bit (MSB) being 1.

[AssortedFantasy]

Its not implemented yet but this is #104