Misleading error message when attempting to perform comptime dereference to pointer whose type doesn't have a well-defined layout when the real pointer value has another type

[gabeuehlein]

Zig Version

0.14.0-dev.2922+f77e1b862

Steps to Reproduce and Observed Output

Currently, the Zig compiler emits a poor-quality error message for the following code:

// a.zig
comptime {
    _ = @as(*const []const u8, @ptrCast(&"foobar")).*;
}

$ zig build-obj a.zig
a.zig:2:52: error: comptime dereference requires '[]const u8' to have a well-defined layout
    _ = @as(*const []const u8, @ptrCast(&"foobar")).*;
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~
referenced by:
    root: lib/std/start.zig:3:22
    comptime: lib/std/start.zig:27:9
    2 reference(s) hidden; use '-freference-trace=4' to see all references

The real problem is that &"foobar" has a type of *const *const [6:0]u8 and that a *const [6:0]u8 can't be coerced into a []const u8 through a pointer dereference.

Expected Output

Something like the following:

a.zig:2:52: error: cannot reinterpret '*const [6:0]u8' as '[]const u8'
    _ = @as(*const []const u8, @ptrCast(&"foobar")).*;
        ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~
a.zig:0:0: note: '*const [6:0]u8' here
    _ = @as(*const []const u8, @ptrCast(&"foobar")).*;
                                        ^~~~~~~~~
a.zig:0:0: note: casted to '*const []const u8' here
    _ = @as(*const []const u8, @ptrCast(&"foobar")).*;
                               ^~~~~~~~
referenced by:
    root: lib/std/start.zig:3:22
    comptime: lib/std/start.zig:27:9
    2 reference(s) hidden; use '-freference-trace=4' to see all references

Of course, the exact error message depends on how type punning at comptime gets implemented (see #9646 for an example). Under that proposal, I'm pretty sure this would be illegal type punning.
Alternatively, I suppose that implicit pointer-to-array to slice coercions like this could be allowed in some way, which would make the above example compile successfully, but that's a topic for another issue.