fix: drop EOL Python 3.8, pin botocore, fix docstring typos, remove dead code

[Hardikrepo]

Summary

This PR fixes several bugs and maintenance issues found in the main branch:

Fix 1 — Remove dead importlib_metadata fallback in __init__.py

The try/except ImportError block around from importlib import metadata was added to support Python < 3.8, but requires-python = ">=3.8" means this fallback is unreachable. Additionally, importlib_metadata is not listed in dependencies, so the fallback would fail at runtime on Python < 3.8 anyway. Replaced with a direct import.

Fix 2 — _handle_ping docstring says "HEAD request" but route is GET

The docstring for _handle_ping() in __init__.py reads "Handle an incoming ping HEAD request", but _ROUTES maps ("GET", "/ping"). SageMaker sends a GET to /ping. Corrected to "GET request".

Fix 3 — Typos in _plugin.py docstrings

Three typos in docstrings:

• ping_fn: "A functions wich indicates" → "A function which indicates"
• output_fn: "seriazizes" → "serializes"
• output_fn: "aplication/json" → "application/json"

Fix 4 — Drop EOL Python 3.8 and fix 3.13-dev in CI

Python 3.8 reached end-of-life in October 2024. It is removed from:

• pyproject.toml: requires-python = ">=3.8" → ">=3.9"
• .github/workflows/test-package.yml: removed '3.8' from matrix
• tox.ini: removed py38 from envlist and [gh-actions] mapping

Additionally, Python 3.13 was released in October 2024, so '3.13-dev' is replaced with '3.13'.

Fix 5 — Add lower bound to unpinned botocore dependency

botocore was listed with no version constraint at all ("botocore"). A fully unpinned dependency can silently pull in incompatible versions. Added botocore>=1.12 as a safe lower bound (the version that introduced the StreamingBody API used in inference_server.testing).

Fix 6 — Remove duplicate test

test_package_has_version and test_version in tests/test_inference_server.py both assert inference_server.__version__ is not None. Removed the duplicate test_package_has_version.

Files Changed

File	Change
src/inference_server/__init__.py	Remove dead importlib_metadata fallback; fix _handle_ping docstring
src/inference_server/_plugin.py	Fix 3 docstring typos
pyproject.toml	requires-python >=3.9; botocore>=1.12
.github/workflows/test-package.yml	Drop 3.8, fix 3.13-dev → 3.13
tox.ini	Drop py38 from envlist and gh-actions mapping
tests/test_inference_server.py	Remove duplicate test_package_has_version

Test plan

• All existing tests pass (tox)
• Linting passes (tox -e linting)
• CI matrix runs cleanly on Python 3.9–3.13
• from importlib import metadata resolves correctly on Python 3.9+