feat: Add protobuf runtime version to x-goog-api-client header

[parthea]

See https://cloud.google.com/apis/docs/system-parameters#definitions
Googlers see b/401039328

This change is built on top of #2386 which fixes a presubmit which is failing at main.

Before:

urlRequestHeaders:
X-Goog-Api-Client: "gl-python/3.9.16 rest/google-auth@2.40.0 gax/2.24.2 gapic/0.0.0"

After (with googleapis/python-api-core#812):

urlRequestHeaders:
X-Goog-Api-Client: "gl-python/3.9.16 rest/google-auth@2.40.0 gax/2.24.2 gapic/0.0.0 pb/6.30.2"

You can test the change by forcing installation of PR googleapis/python-api-core#812 using the change below

py3133partheniou@partheniou-vm-3:~/git/gapic-generator-python$ git diff noxfile.py
diff --git a/noxfile.py b/noxfile.py
index 117ca47c..899fdd4a 100644
--- a/noxfile.py
+++ b/noxfile.py
@@ -373,6 +373,10 @@ def showcase_library(
"--upgrade",
"google-auth[aiohttp]!=2.40.0",
)
+ session.install(
+ "--ignore-installed",
+ "google-api-core @ git+https://github.com/googleapis/python-api-core.git@add-protobuf-runtime-header",
+ )
else:
# The ads templates do not have constraints files.

and then run nox -s showcase_w_rest_async-3.9

[vchudnov-g]

Note that the implementation in https://github.com/googleapis/python-api-core/blob/118bd96f3907234351972409834ab5309cdfcee4/google/api_core/client_info.py#L114 is problematic. Not only does it depend on us constructing the format string in the same order as the __dict__ entries, but it also requires these guard statements at the calling site. We can/should refactor that code to make this unnecessary going forward, except we'll still need to account for people using api_core from before the refactor.....we'll be stuck with these guard clauses ~indefinitely now.

[parthea]

Correct, the format string has a fixed order but it's not clear how problematic this is.

but it also requires these guard statements at the calling site

The guard statements are often necessary for backwards compatibility reasons for any new feature that we add to google-api-core

[vchudnov-g]

Yeah, ignore my comment about the order. I got confused.

But this is something that shouldn't need a guard clause, if api-core had been implemented differently. IMO, we should be able to set arbitrary key-value pairs in ClientInfo, without the need for the guard clause here. When emitting the header, ClientInfo would only look for the keys it knows about to create the format string. So older versions would not use the protobuf info, but as soon as people were using the newer version of api-core, that would become part of the header. This would make things more extensible, though admittedly it would be less type-safe, since a typo would cause the header token to be omitted.

Oh, well. Hindsight.