bug: ignore non-deployment entries in the project's live directory

[Amadeus-22]

Closes #676.

• With Paper Trading (or any brokerage that supports an initial cash balance or holdings), lean live deploy and lean cloud live deploy aborted with time data '.DS_Store' does not match format '%Y-%m-%d_%H-%M-%S' when <project>/live contained an entry that isn't a deployment timestamp, like a Finder .DS_Store or a directory from an earlier --output.
• _get_last_portfolio now skips the entries whose name doesn't parse, as is_backtest_output_directory in optimize.py already does. Deployment directories are read as before.
• Added a parametrized test (a .DS_Store file and a custom directory next to a real deployment) that runs lean live deploy. It fails on master and passes with the fix.
• pytest tests/commands/test_live.py tests/commands/cloud/live passes (1553 passed, 18 skipped, Python 3.12) and python static_analysis.py is clean.

🤖 Generated with Claude Code

[jhonabreul]

Thanks for the fix. I pushed a follow-up commit (d2c9a37) on top of it.

Cases that still failed with the previous commit

<project>/live entry	Result
my-run, from --output live/my-run	Skipped, so an older cloud deployment won and its portfolio was offered
1960-01-01_00-00-00	Parses, but .astimezone() ran outside the try and raises OSError on Windows

The test could not catch these: its deployment directory was empty, so _get_last_portfolio returned None whether or not it was read.

What changed

• Timestamp names convert to UTC inside the try. Any other directory with an L-*.json is a deployment dated by that file's mtime.
• The name parser moved to output_config_manager.get_output_directory_timestamp, shared by optimize.py and live_utils.py. The format is now OUTPUT_DIRECTORY_NAME_FORMAT.
• Ignored entries are logged with --verbose.
• tests/components/util/test_live_utils.py tests _get_last_portfolio directly with a state file. The command test seeds one too and asserts the cash balance is offered.
• The custom-directory and pre-1970 cases fail on the previous commit. The live, optimize and backtest suites pass (1709 passed, 18 skipped) and static_analysis.py is clean.

The elif branch still picks the state file by mtime from the working directory, which can differ from the directory that won the comparison. That predates this PR.

[jhonabreul]

Pushed 96ae80c to fix CI: the 1960-01-01_00-00-00 case only fails to convert on Windows, so on Linux and macOS it is an older deployment and nothing is logged for it. The test now only asserts that the newer deployment's state is still returned.

[jhonabreul]

Pushed eb2f4ea for the Python 3.9 failure: mock.patch with a dotted path resolves lean.commands.live to the click group there, since lean/commands/__init__.py imports the group under the package's name. The test now patches the module object instead. Verified on 3.9 and 3.14.

[Amadeus-22]

Thanks for the follow-ups, @jhonabreul. I read d2c9a37, 96ae80c and eb2f4ea: matching only the timestamped deployment entries covers the --output live/ case I missed, and the notes on the Windows-only timestamp conversion and the Python 3.9 mock.patch behavior are useful for the next tests I write.