Malformed packed varints read beyond field length

With the following protobuf

syntax = "proto3";

message RepeatedBug {
  bytes name = 1;
  repeated int32 values = 2;
  int32 id = 3;
  bytes should_never_appear = 4;
}

And the following (malformed) input:

12       // (tag 2 | Length Delimited)
01       // Length of 'values'
80       // *** Invalid truncated varint ***
18       // (tag 3 | Varint)
22       // ID value 34, which happens to be equivalent to (tag 4 | Length Delimited)
0A       // (tag 1 | Length Delimited), which happens to be equivalent to a length of 10
09       // Length of 'name'
313233   // UTF-8 Value "123456789"
343536
373839

(which can be generated by this convenient protoscope):

2:LEN {`80`}
3:VARINT 4:LEN
1:LEN {"123456789"}

Wire-Swift treats the 3:VARINT as part of the length-delimited field, and produces RepeatedBug(values: [3072], should_never_appear: "\u{09}123456789"), instead of rejecting the input as malformed. (Most instances of this bug would result in an unexpected error somewhere else, as the input stream has been desynchronized, but I managed to find this one that puns to a different "valid" structure instead.)

As far as I can tell the Kotlin implementation has the same bug, but I didn't test it. Similarly, both implementations seem vulnerable to the same issue when it's a nested message that's truncated, rather than a packed field. (It would not surprise me if these were covered by Google's protobuf conformance tests, but I didn't check that either.)

Start with the Wire-Swift and Kotlin protobuf decoding implementations and reproduce the malformed packed-varint input from the issue. Compare their handling with the linked protobuf conformance tests, including truncated nested messages; done means malformed input is rejected without consuming bytes from the following fields.