Add persistent sqlite3-backed API metadata cache

[bendichter]

Summary

• Add APIMetadataCache class in dandi/apicache.py — a persistent sqlite3-backed cache for API metadata responses, keyed by (api_url, entity_type, entity_id) and validated against modified timestamps
• Add cache: bool = False parameter to DandiAPIClient.__init__() to opt in to caching
• Integrate cache into RemoteDandiset.get_raw_metadata() and BaseRemoteAsset.get_raw_metadata() — no extra API calls needed for staleness checks
• Respects existing DANDI_CACHE env var ("ignore" disables, "clear" wipes)
• Add module reference docs (apicache.rst), caching section in dandiapi.rst, and updated DEVELOPMENT.md

Test plan

• python -m pytest dandi/tests/test_apicache.py -v — 8 unit tests covering miss, hit, staleness, update, clear, different entity types, DANDI_CACHE=ignore, and DANDI_CACHE=clear
• pre-commit run --files dandi/apicache.py dandi/dandiapi.py dandi/tests/test_apicache.py — all checks pass
• Manual: python -c "from dandi.dandiapi import DandiAPIClient; c = DandiAPIClient('https://api.dandiarchive.org/api', cache=True); ds = c.get_dandiset('000027'); print(ds.get_raw_metadata()['name'])" — run twice, second should be faster

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 87.87879% with 16 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.14%. Comparing base (0fe497d) to head (f5a38de).

Files with missing lines	Patch %	Lines
dandi/dandiapi.py	56.25%	14 Missing ⚠️
dandi/apicache.py	95.83%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1815      +/-   ##
==========================================
+ Coverage   75.11%   75.14%   +0.02%     
==========================================
  Files          84       86       +2     
  Lines       11925    12050     +125     
==========================================
+ Hits         8958     9055      +97     
- Misses       2967     2995      +28     

Flag	Coverage Δ
unittests	75.14% <87.87%> (+0.02%)	⬆️

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

☔ View full report in Codecov by Sentry.
📢 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]

does it work ok on large ones like 000026? I just wonder that may be sqlite3 can quickly become inefficient for some uses. E.g. how large it would become after listing all dandisets and would it still perform ok? as should there be some per-dandiset cache may be instead?

[yarikoptic]

frankly I am not sure yet if we can rely on modified on per asset, see e.g. a random first hit

• zarr might change without having "modified" adjusted for it dandi-archive#1432
  but also can dig deeper to still open
• Dandiset "modified" timestamp not updated when entries deleted from Zarr dandi-archive#1871
  etc :-/

and we had touched on those prior in

• caching of asset metadata #1670