fix(plugins): clean up standalone sender ty warnings

[GodsBoy]

Summary

• Rebase the stale fork branch onto current main while preserving the original focused typing cleanups
• Use importlib for the optional aiohttp import in Google Chat standalone send without changing its runtime behaviour
• Narrow the optional Google Chat service_account module in both standalone-send credential mock tests
• Annotate IRC setup defaults so ty does not infer a heterogeneous value union for PlatformEntry keyword arguments
• Record the review follow-up and verification contract in docs/plans/2026-07-13-001-fix-pr-24210-review-followup-plan.md

Verification

• scripts/run_tests.sh tests/gateway/test_google_chat.py tests/hermes_cli/test_setup_irc.py - 168 passed
• ruff check . - passed
• python scripts/check-windows-footguns.py --all - passed
• python -m compileall -q plugins/platforms/google_chat/adapter.py tests/gateway/test_google_chat.py tests/hermes_cli/test_setup_irc.py - passed
• CI-style Ruff and ty diff against the rebased origin/main - no new Ruff issues and 22 fewer ty diagnostics
• Full local suite - 40,402 passed and 23 failed in unrelated, unchanged areas; the corresponding failure set reproduced on current main, apart from one timing-sensitive cache invalidation test in unchanged code

Addresses the advisory ty diff reported on #22461: #22461 (comment)

[GodsBoy]

CI triage note: the current red checks appear to be inherited from main, not caused by this PR.

This PR only changes:

• plugins/platforms/google_chat/adapter.py
• tests/gateway/test_google_chat.py
• tests/hermes_cli/test_setup_irc.py

Relevant PR-specific checks passed:

• ruff + ty diff ✅
• ruff enforcement (blocking) ✅
• targeted local tests for the touched Google Chat / IRC setup files passed before opening this PR

Comparison against latest main at 99ad2d137:

• Windows footguns (blocking) fails on both PR and main with the same tools/process_registry.py:588 os.killpg(..., signal.SIGKILL) findings.
  • PR job: https://github.com/NousResearch/hermes-agent/actions/runs/25717145318/job/75509539578
  • main job: https://github.com/NousResearch/hermes-agent/actions/runs/25716556994/job/75507694576
• e2e fails on both PR and main with the same 9 tests/e2e/test_platform_commands.py /new session reset failures.
  • PR job: https://github.com/NousResearch/hermes-agent/actions/runs/25717145376/job/75509539756
  • main job: https://github.com/NousResearch/hermes-agent/actions/runs/25716556968/job/75507694372
• test fails broadly on both PR and main; the PR has no failures outside the set already observed on main (main currently has one extra unrelated failure in tests/gateway/test_approve_deny_commands.py).
  • PR job: https://github.com/NousResearch/hermes-agent/actions/runs/25717145376/job/75509539804
  • main job: https://github.com/NousResearch/hermes-agent/actions/runs/25716556968/job/75507694361

I am intentionally not broadening this cleanup PR to fix unrelated CI/main breakage. The actual cleanup in this PR is scoped to the advisory ty warnings reported on the already-merged #22461: #22461 (comment)

[teknium1]

Thanks for the focused ty cleanup. The three original source patterns remain on current main, including the optional aiohttp import at plugins/platforms/google_chat/adapter.py:3250 and the heterogeneous PlatformEntry(**defaults) setup at tests/hermes_cli/test_setup_irc.py:19-39.

Problems

• The Google Chat test narrowing only covers the API-failure test. Current main's preceding refresh-success test still dereferences _gc_mod.service_account.Credentials at tests/gateway/test_google_chat.py:2858, :2859, and :2874, leaving the same potentially-None typing pattern outside this PR's hunk.

Suggested changes

• Apply the same local service_account binding and non-None assertion to that refresh-success test before salvaging this cleanup.

Automated hermes-sweeper review.