fix schema-serde sparse collection deser to not point to the same value

[lucix-aws]

Found this while working on a new body of protocol tests with the Smithy team. Fortunately no released services that are on schema-serde in the downstream SDK are affected.

Schema-serde deserializers for sparse lists/maps declared the scratch variable once outside the read loop and appended &vv for every element, so every non-null element aliases the same variable and ends up as the last decoded value. Nulls are fine (appended as literal nil) but the non-null entries get overwritten.

["foo", null, "bar"]      -> ["bar", nil, "bar"]
{"a":1, "b":null, "c":3}  -> {"a":3, "b":nil, "c":3}

The existing sparse deserialize tests only ever have a single null element so this was not caught (awsJson1_1/null.smithy):

{
    id: "AwsJson11SparseListsDeserializeNull"
    protocol: awsJson1_1
    code: 200
    body: "{ \"sparseStringList\": [null] }"
    params: { sparseStringList: [null] }
}

Caught this with a new test like so:

{
    id: "AwsJson10SparseListOfScalarsWithNullsResponse"
    protocol: awsJson1_0
    code: 200
    body: "{ \"strings\": [\"foo\", null, \"bar\"], \"integers\": [1, null, 3] }"
    params: { strings: ["foo", null, "bar"], integers: [1, null, 3] }
}

[lucix-aws]

Sidebar, I would honestly like to see if we can not generate these, and instead have them as generics in the runtime somehow. But this PR is not the place for that.