fix: send list paging parameters in the JSON body

[marevol]

Summary

fessctl <resource> list --size N --page M had no effect against Fess 15.8 and 15.9. The admin list endpoints read size, page and the endpoint's filters from the JSON body of the GET request and ignore the query string, but FessAPIClient.send_request only sent params=.

Observed with GET /api/admin/<resource>/settings and size=1&page=2:

Endpoint	Fess	Query string only	JSON body only	Both
scheduler	15.9.0-SNAPSHOT	15 rows (total 15)	1	1
role	15.9.0-SNAPSHOT	2 rows (total 2)	1	1
labeltype	15.9.0-SNAPSHOT	3 rows	1	1
scheduler	15.8.0	14 rows	1	1
role	15.8.0	2 rows	1	1
labeltype	15.8.0	3 rows	1	1

So every list command returned the server default of 25 rows. A script that pages with --page 1,2,... until it has total rows (for example a migration helper) read page 1 repeatedly and silently skipped the rest once there were more than 25 entries.

Change

• Action.LIST requests now go through httpx.request("GET", ..., params=params, json=params): the parameters travel in the JSON body (what the server reads) and stay in the query string (harmless; keeps a server that reads either channel working). One place in send_request, not the 23 list_* methods.
• Action.GET (single record by id) is unchanged.
• Visible effect: the existing --size default of 100 now actually applies.

Fess 14.x was not probed (no image at hand). The client already parses the Fess version if a gate is ever needed.

Tests

• Unit (tests/unit/test_client.py): TestListPaging asserts that all 23 list_* methods put page/size into the JSON body and the query string, that a list call without parameters sends no body, and that a GET by id sends no body. The new list tests fail on the previous client (25 of 26) and pass now; two existing tests that mocked httpx.get for list calls now mock httpx.request. uv run pytest tests/unit: 155 passed.
• Integration (tests/commands/test_scheduler.py::test_scheduler_list_size_and_page): --size 1 returns exactly one row on page 1 and a different row on page 2, with total greater than 1. It failed before the change (15 == 1 on 15.9, 14 == 1 on 15.8.0) and passes on both versions.
• tests/commands against 15.8.0: 47 passed, 1 skipped (test_tagtype, 15.8.0 has no tagtype API). Against the 15.9 snapshot: 47 passed, 1 failed - test_user_crud_flow at the update step ("roleA is not a registered role or group id"); it fails the same way without this change.