AddNullMemberInitializers should check if a member is assigned before instantiating with ??= new()

[TimothyMakkison]

Is your feature request related to a problem? Please describe.
When mapping paths, MapProperty initializes parents of members using ??= new() to prevent exceptions. This doesn't check if the member has already been set resulting in redundant code.

[MapProperty("NestedValue", "Nested.Value")]
[MapProperty("Parent", "Nested")] 
partial B Map(A source);

class A { public string NestedValue { get; init; } public C Parent { get; set; } }
class B { public C? Nested { get; set; } }
class C { public string Value { get; set; } }

// Generates
var target = new global::B();
target.Nested ??= new(); // redundant
target.Nested = source.Parent;
target.Nested.Value = source.NestedValue;
return target;

// Should generate
var target = new global::B();
target.Nested = source.Parent;
target.Nested.Value = source.NestedValue;
return target;

Describe the solution you'd like
AddNullMemberInitializers should check if the member is already set before creating a MemberNullAssignmentInitializerMapping.

[TimothyMakkison]

Just realised that the nullable initializer is required. Even if the source is known to not be null the targets setter could make the property null (very unlikely).

Instead the nullable initializer should be added after the assignment as it's currently redundant in these cases.

// Generates
var target = new global::B();
target.Nested = source.Parent; // nested setter could do something weird (rare)
target.Nested ??= new(); // should be after
target.Nested.Value = source.NestedValue;
return target;

I don't know if it's worth accounting for bizarre setters. WDYT

[latonz]

Is it really required? Isn't it only required when source.Parent is nullable? Otherwise the same problem would exist with target.Nested ??= new(); (a setter could assign null here too), but I wouldn't account such weird setters.

[TimothyMakkison]

Good point, no idea why I didn't think of that 😆