fix(docs): correct stale h5ad examples in cli.py docstrings

[iamcipherdev]

Closes #38

Several Examples: blocks in src/adata/cli.py still used the pre-rename h5ad command name, and export sparse / export image showed a positional output that the commands reject (output goes to -o or stdout). All six blocks fixed per the issue.

Also extended tests/test_docs_are_accurate.py: it now walks the real Click tree and validates every adata ... example line in the --help text (real command + real options, including short flags). I verified the new check fails on the pre-fix h5ad lines and passes after; 283 tests pass in the two touched test files.

[Claptar]

Thanks for this. The docstring fixes look right, and a test that walks every --help example is a good guard to have.

One non-blocking note. test_help_examples_name_real_commands_and_options checks that the command path resolves and that each option is real. It doesn't check the number of positional arguments, though. So an example like adata export sparse data.h5ad X matrix.mtx, with a stray positional output, would pass as long as it starts with adata. That was half of what #38 reported.

Click can do that check for you. For each example, resolve the command, then call command.make_context(name, args) with the remaining tokens. That raises click.UsageError on an unexpected extra argument, a missing required one, or an unknown option, so it could also replace the hand-rolled option check. Happy to take that as a follow-up if you'd rather keep this PR as it is.