frontend: type [] from the other initializers of its name

[notactuallytreyanastasio]

let xs = items ?? [] with items: List<String>? typed xs as
List<AnyValue>. The ?? macro expands to an if that assigns []
on one branch and items on the other, so xs has two initializers.
The typer typed each with no context, [] came out List<AnyValue>,
and the union of that with List<String> is List<AnyValue>. That
is a legal Temper type, but java, rust and cpp then declare the local
with it and assign a List<String>:

List<Object> xs__3;
...
xs__3 = items__1;   // incompatible types

TyperPlan already described this case (h = [] on one branch,
h = ["foo"] on the other) and left a TODO to exclude initializers
that need context. This does that, the way panic() initializers
are already handled: when a name has an empty list literal among its
initializers and at least one other that is not null, the [] is
left out of the initializers, moved to the end of the type order, and
typed with the name's inferred type as its context. The java output
is now

List<String> xs__3;
if (items__1 == null) {
    xs__3 = List.of();
} else {
    xs__3 = items__1;
}

and the case prints 2 and 0 on js, py, lua, java, rust and cpp.

A null initializer does not count as the other one: it only says the
type is nullable, so if (b) { zs = null } else { zs = [] } still
types zs as List<AnyValue>?, as before. A name whose only
initializers are [] is unchanged too. This only covers the empty
literal; a non-empty literal of a narrower type on one branch still
widens the union, which is #548's territory. csharp was not run.

Fixes #547.

Repro: https://github.com/notactuallytreyanastasio/temper-issue-repros/tree/main/25-coalesce-empty-list-any-value

🤖 Generated with Claude Code