Write uint32 values as varints of at most 5 bytes

[merlimat]

Follow-up to #42: reuses the writeRawVarUInt() it added.

Problem

LightProto sized and wrote a uint32 like an int32. A value >= 2^31, which is negative as a Java int, was therefore written as the 10-byte sign extension. protobuf writes a uint32 as an unsigned 32-bit varint, at most 5 bytes.

Every reader decodes both encodings to the same value, so this costs 5 extra bytes per such value. It also means protobuf-java's own encoding of these values re-serializes to a different size than it was parsed from (#43 covers the size cache side of that).

Example

NumbersOuterClass.Numbers.newBuilder().setXUint32(0x80000000).build().getSerializedSize();  // 6
new Numbers().setXUint32(0x80000000).getSerializedSize();                                  // 11

The current code (LightProtoNumberField):

} else if (field.getProtoType().equals("int32") || field.getProtoType().equals("uint32")) {
    writer = "writeRawVarInt";                                   // a negative int takes 10 bytes
...
} else if (field.getProtoType().equals("uint32")) {
    return String.format("LightProtoCodec.computeVarIntSize(%s)", name);   // 10 for a negative int

Change

uint32 is now sized with computeVarUIntSize() and written with writeRawVarUInt():

} else if (field.getProtoType().equals("int32")) {
    writer = "writeRawVarInt";
} else if (field.getProtoType().equals("uint32")) {
    writer = "writeRawVarUInt";
...
} else if (field.getProtoType().equals("uint32")) {
    return String.format("LightProtoCodec.computeVarUIntSize(%s)", name);

The output now matches protobuf-java for singular, repeated, packed and map uint32 fields. This changes LightProto's output bytes for values >= 2^31. Parsing is unchanged, and both encodings decode the same everywhere.

Testing

NumbersTest and RepeatedNumbersTest cover 2^31 and 2^32 - 1. For singular, repeated and packed fields, sizes and bytes match protobuf-java. These tests fail without the change. Full mvn verify passes.