GH-139: [Flight] Stop return null from MetadataAdapter.getAll(String) and getAllByte(String)

[axreldable]

What's Changed

CallHeaders has 3 implementations:

• FlightCallHeaders
• ErrorFlightMetadata
• MetadataAdapter

Before this change:
MetadataAdapter could return null from getAll(String) and getAllByte(String) when there were no values for the key, because gRPC’s Metadata.getAll() returns null in that case. This was undocumented and forced callers to null-check.

After this change:
All 3 implementations consistently return an empty iterable (never null) when the key is absent or has no values. The contract is documented on the interface and covered by tests for each implementation.

---

This contains breaking changes.
MetadataAdapter.getAll(String) and getAllByte(String) return empty iterator instead of null.

---

Closes#139.

[axreldable]

Please add the enhancement label.

[ennuite]

Hello @axreldable
This looks good to me, thank you for it.

I note that this also fixes an NPE in MetadataAdapter.getAllByte(String key)

[axreldable]

Hello @axreldable This looks good to me, thank you for it.

I note that this also fixes an NPE in MetadataAdapter.getAllByte(String key)

Correct, it fixes NPE for getAll(key).spliterator() in MetadataAdapter.getAllByte(String key) method.

[jbonofre]

Overall LGTM.
Can you just clarify the case when it is null in the ServerSessionMiddleware please ?

[axreldable]

@jbonofre , some CI pipelines failed. Do you think it relates to the changes or the pipelines are flaky?

[jbonofre]

@axreldable let me take a look. I will get back to you.

[jbonofre]

The verify job failed to flaky test (memory allocation). I ran again the jobs to double check.

[axreldable]

Thank you, @jbonofre ! Looks like tests passed this time.