feat: add Assets/Computer and Management/Contract endpoint support

[baraline]

Adds GLPI asset and contract endpoint support to the client, with an agent
skill for each family. 28 new public methods, additive throughout — no
existing signature or behaviour changes.

What's in scope

Assets/Computer is the only asset itemtype implemented, of the 24 the API
exposes. That was a deliberate choice: prove the shape on one itemtype, with
its documentation, skill and tests, before repeating it 23 times. The
contract linkage is the part most likely to generalise badly, so it is
included rather than deferred.

• Assets/Computer — flat CRUD plus search/iter_search.
• Computer-to-contract links — the Contract_Item join, exposed as
  list_/get_/link_/update_/unlink_computer_contract. Per-type method names
  rather than a generic list_asset_contracts, so the second asset type
  can't silently inherit a signature that doesn't fit it.
• Management/Contract — full CRUD plus the Cost sub-resource.
• Dropdowns/ContractType.
• Skills — glpi-asset-workflow and glpi-contract-workflow.

The client stamps the link's itemtype itself on both the create and update
paths, because the API types that field as a free string and will happily
store Monitor on a computer's link. There's a test that passes a
deliberately wrong itemtype to prove the override holds.

Verification

The gate: 1555 tests, ruff, mypy --strict, unasync_build.py --check, and a
zero-warning Sphinx build.

All 20 models were checked field-for-field against the OpenAPI document, and
then against a live GLPI 11 instance — this repo's history is largely a
record of the two disagreeing, so the document alone isn't evidence. Diffing
each served payload against its model across Computer, Contract,
ContractCost, Contract_Item and ContractType found nothing served that isn't
modelled. Three behaviours were measured and are now pinned as tests:

• Contract.alert is validated by nothing. The document's enum lists
  64/72 where its own prose numbers the same meanings 16/24. The server
  stores all eight of 0/4/8/12/16/24/64/72 unchanged and rejects none, so
  neither listing describes an enforced set. Modelled as a plain int — an
  enum would reject values the server accepts, and would have to pick one of
  two contradictory numberings to name them by.
• force=True genuinely hard-deletes even though this client sends it in
  the request body while the document declares it a query parameter. Every
  teardown in the suite rests on this; a server that ignored the body would
  leave soft-deleted records accumulating while still reporting success.
• Computer.entity carries completename and Contract.entity does
  not, so they use different reference types.

Contract.date_begin is datetime.date, not datetime, so it stays out of
the server-clock conversion that rewrites aware timestamps — on a date-only
field that could roll the value to the neighbouring day.

Reviewer notes

• The integration tests skip silently without credentials, so they will
  not run in CI. They were run locally against preprod; to reproduce, run
  pytest integration_tests/ -m integration from a checkout with secrets/
  populated and confirm the count rather than trusting a green result.
• _sync/ is generated. Review _async/ and let unasync_build.py --check
  cover the rest.
• Known and accepted, not fixed: the three new client modules have no
  FailingTransportRecorder tests (raise paths are covered generically), and
  the list_* helpers default to limit=50 without saying so in the
  signature.

🤖 Generated with Claude Code

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.59%. Comparing base (0d43528) to head (763f709).
❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #38      +/-   ##
==========================================
+ Coverage   97.31%   97.59%   +0.28%     
==========================================
  Files          80       90      +10     
  Lines        2826     3163     +337     
==========================================
+ Hits         2750     3087     +337     
  Misses         76       76              

☔ 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.