fix: sort LicenseType members by value for reproducible schemas

[candleindark]

Fixes #433.

Problem

LicenseType is built by iterating Config.licenses, which is a set. The order of iterating a set of Enum members depends on PYTHONHASHSEED, which CPython randomizes per process, so the member order of LicenseType varied from process to process.

That member order is what fixes the order of the elements in the enum array of the LicenseType definition under $defs, which is emitted into all four generated JSON Schemas (dandiset.json, published-dandiset.json, asset.json, published-asset.json). Two invocations of tools/pubschemata.py on the same commit could therefore emit different files.

Fix

Sort the members by value when building the enum. Config.licenses stays a set, with the sort applied at the single point where the order becomes observable. Validation behavior is unchanged, since JSON Schema enum is order-insensitive.

Merge order

Recommend merging to master before #419 is released. This PR carries no DANDI_SCHEMA_VERSION bump, since #419 already bumps it to 0.8.0. The fix settles the order on ["spdx:CC-BY-4.0", "spdx:CC0-1.0"], which differs from what dandi/schema releases/0.7.0/ holds, so merging ahead of #419's release lets 0.8.0 be published once with the sorted order.

Test plan

• LicenseType member order is identical across PYTHONHASHSEED in {0, 1, 2, 3, 4, 5, 42, 1234}
• tools/pubschemata.py output is byte-identical between seeds 0 and 3; before the fix, all four schema files differed
• pytest dandischema: 258 passed, 19 skipped
• pre-commit and mypy clean

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 47.48%. Comparing base (cd26d54) to head (0114c13).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #434   +/-   ##
=======================================
  Coverage   47.48%   47.48%           
=======================================
  Files          19       19           
  Lines        2367     2367           
=======================================
  Hits         1124     1124           
  Misses       1243     1243           

Flag	Coverage Δ
unittests	47.48% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[yarikoptic]

Suggested change

	# The `sorted()` call is required, not cosmetic. `Config.licenses` is a `set`,
	# and the order of iterating a `set` of `Enum` members depends on
	# `PYTHONHASHSEED`, which CPython randomizes per process. Without sorting, the
	# member order, and hence the `enum` array of every JSON schema generated from
	# these models, varies from run to run. Pydantic emits member values rather
	# than names into the `enum` array, which is why `value` is the sort key.
	# `sorted()` since `Config.licenses` is a `set`.

is likely the absolute maximum needed here, the rest is too AIsloppy to my taste, or am I missing some detail which might be crucial if omitted?

[candleindark]

I wanted to point out why ordering is needed here not that Config.licenses is unordered. I agree that the original one is too wordy. I have shorten it to just one line. Additionally, I have removed the change to the following docstring.

Please take another look.