False errors for maybe operator/optional chain

[Lexikos]

a := {123:IsObject, 222:{__item:42}}
assert (a?.123) ?? 0 = IsObject
; this?.1 can be member access, but in this case is ternary,
; because a?.123() is always member access:
assert (this?.1 && a?.123() : 0) ?? 0
assert ((this?.1 && a?.222[] : 0) ?? 0) = 42

f() => 42
assert f.%unset%() = 42
assert f.%a?%() = 42
a := 'Bind'
assert f.%a?%() is BoundFunc

assert(c) => c ; (Not the actual definition)

This is taken from some test units. With v2.4.6, there are some incorrect "Missing ':'" diagnostics:

• (a?.123). This should be interpreted as member access like a.123, because it can't be ternary. I think the exact rule isn't documented, but basically it just tries to make the new syntax work without breaking any scripts (so ternary takes precedence).
• this?.1. The comment explains, but I forget how it actually works, so feel free to ignore these two.
• f.%a?%. This just allows a to be unset; "unset resolves to the default property (either __Item or Call)."

I think that it would be safe for the extension to assume ?. is member access if the colon is missing, although the actual interpretation might differ in some complicated cases.