Make make lint able to fail, and match the lint workflow

[mrubash1]

make lint could not fail. Makefile line 3 was ruff check --exit-zero ., and --exit-zero forces an exit status of 0 regardless of findings, so a green make lint predicted nothing about CI. It also never ran ruff format --check, though .github/workflows/lint.yml does.

This makes the target run the same three checks as the workflow, in the same order.

The two failure modes, demonstrated

Both were checked against the current tree with a deliberately broken file, using CI's pinned versions (ruff 0.1.6, snakefmt 0.8.5) under Python 3.9:

broken file	before	after
unused import (F401)	prints Found 1 error., exits 0	exits 1
lint-clean but misformatted	no output at all, exits 0	exits 1

The second row is the worse of the two. --exit-zero at least still printed the finding, so someone watching the scrollback could notice it. But with no ruff format --check at all, a file that CI rejects produced completely clean local output and a zero exit — there was nothing to notice.

Also: the order of the two commands in make format

make format ran ruff format and then ruff check --fix. The fixer runs last, so its edits are never formatted afterwards. That is invisible while make lint ignores formatting, but this PR makes make lint enforce it, so it would become a fixed-point violation introduced by this change: run make format, then make lint fails.

It is reachable with an ordinary edit. Given this format-clean file:

import os

import sys


def f():
    return sys.path

ruff check --fix removes the unused os and leaves the blank line that followed it, so the file starts with a blank line and ruff format --check then rejects it. Reversing the two commands produces a clean tree from the same input. This also matches ruff's own guidance to run the fixer before the formatter.

Verification

• make lint exits 0 on the current tree.
• make lint exits non-zero for a lint error and for a formatting error (the table above); a fix that cannot be shown to fail is not verified.
• make format rewrites nothing on the current tree — the only modified file afterwards is the Makefile itself — and still fixes both broken files above.

What was not verified

make test does not pass on this branch, and does not pass on unmodified main either. Two rules fail:

rule leiden_clustering:  ImportError: Matplotlib requires numpy>=1.25; you have 1.23.5
rule dim_reduction:      ModuleNotFoundError: No module named 'pkg_resources'

That is the envs/analysis.yml dependency drift fixed in #101, reproduced here independently on a clean checkout of current main with a fresh conda solve. It is unrelated to this PR, which touches only the lint and format targets in the Makefile. Worth noting as a second data point for #101: the drift is not specific to one machine or one day's package index.

make smoke-test was not run because that target does not exist on this base; it arrives with #101.

Stacking: #101 also edits the Makefile, adding a smoke-test target further down. The hunks are far apart and were checked to apply cleanly in either order.