Add snmpd system tests with Valgrind and fix memory leak

[SoundGoof]

Summary

Add an end-to-end system-test suite that runs the public Python API against a real snmpd instance.

The suite runs normally and under Valgrind on every pull request using ubuntu-latest.

Changes

• Add a GitHub Actions workflow for pull requests and pushes to master
• Start an isolated SNMP agent on 127.0.0.1:1161
• Test SNMPv1 and SNMPv2c operations
• Test both convenience functions and Session methods
• Cover GET, GETNEXT, GETBULK, SET, and WALK
• Cover multiple varbinds, error paths, and session lifecycle
• Add Varbind and VarList regression tests
• Add a Podman environment using ubuntu:latest
• Run all discovered tests in one Valgrind process
• Fail CI on definite memory leaks or Valgrind errors

Memory fixes

The new Valgrind coverage found and fixed two leaks in netsnmp_walk():

• Removed a duplicate OID buffer allocation that overwrote the original pointer
• Freed the final unsent GETNEXT PDU when a walk completes

Verification

Tested locally with Podman using Ubuntu 26.04 LTS and Python 3.14:

Ran 31 tests ... OK
Ran 31 tests under Valgrind ... OK

definitely lost: 0 bytes
indirectly lost: 0 bytes
possibly lost: 0 bytes
ERROR SUMMARY: 0 errors

[bluecmd]

Lgtm but fix the awk. It's fine that the tests don't pass, we will merge it and work on fixing the issues one by one