Add to existing protobuf namespace instead of overwriting it

[taylorcode]

Addresses a bug related to loading multiple pbjs-generated JS files on one page.

When loading .js files generated by pbjs (static-module) if those .js files contain protos that belong to the same namespace, they will clobber each other at runtime.

Repro

a.proto

syntax = "proto3";
package my_pkg;

message MyMessageA {
    string a_field = 1;
}

b.proto

syntax = "proto3";
package my_pkg;

message MyMessageB {
    string b_field = 1;
}

pbjs -t static-module -w es6 -o a.js a.proto

pbjs -t static-module -w es6 -o b.js b.proto

After running these commands, a.js and b.js both contain:

export const my_pkg = $root.my_pkg = (() => {

So if b.js is loaded after a.js, it will erase the entries created by a.js, and vice versa.

Solution
This PR makes a change to code generation so the protobuf namespaces are extended at runtime:

export const my_pkg = $root.my_pkg = ((my_pkg) => {
   ...
})($root.my_pkg || {});

This changed solved the problem for us at dropbox.

[alexander-fenster]

Hi @taylorcode, would it be possible to have a test that covers this code generation change?

[taylorcode]

@alexander-fenster I regenerated the fixtures with node scripts/gentests.js and added a test, but I'm now skeptical that this is the right approach.

While writing the test I discovered that this is not an issue when --root is configured to be unique for every file; this only occurs when default setting is used since it adds all new packages as properties to a $protobuf.roots["default"].

[alexander-fenster]

@taylorcode Yes, using different roots is what we do in Google Cloud client libraries. It still could make sense not to overwrite root though. Let us think about it a little bit.