fix(upload): keep queued ingest names within filesystem and processor limits

[AlfonsoDehesa]

Problem

Browser uploads of long file names fail with:

Failed to queue upload for processing

and the CWA log records OSError: [Errno 36] File name too long. This
hits real-world names from sources like Anna's Archive dumps that carry
title, series, publisher, ISBN, hash, and source information all in the
filename.

Root cause

_get_ingest_path in cps/editbooks.py only knows about the final
basename. Three sibling names share that stem in the ingest directory:

• the streaming .uploading temp file
• the .cwa.json sidecar manifest written for add-format uploads
• the .cwa.failed.json copy that scripts/ingest_processor.py
  preserves on add-format failure

The destination filesystem rejects any of those once the stem + suffix
exceeds its per-component name limit. scripts/ingest_processor.py
also silently truncates queued basenames over 150 characters before
looking up their manifest, which would orphan the sidecar for long
add-format uploads even on filesystems with a generous name limit.

Fix

Allocate one budget that respects both limits at the same time:

• Query the ingest directory's PC_NAME_MAX via os.pathconf, with a
  documented 255-byte Linux/POSIX fallback when the call is unavailable
  or returns an invalid value.
• Reserve the longest suffix actually created around an upload:
  .uploading, .cwa.json, .cwa.failed.json.
• Cap the queued basename at CWA's ingest processor 150-character
  threshold so the processor never renames it after we hand it off.
• Truncate only the stem, on a UTF-8 boundary, preserving the sanitized
  extension and the uniqueness prefix.
• Keep ordinary short filenames byte-for-byte unchanged.
• Raise a clear ValueError when the fixed components alone exceed the
  budget.

Tests

tests/unit/test_editbooks_ingest_path.py covers the new behaviour
end-to-end:

• new-book and add-format prefixes
• ordinary short names preserved byte-for-byte
• multibyte stems and extensions
• every sidecar path fitting the filesystem limit
• the 150-character processor cap
• pathconf returning OSError / 0 / -1 / None / non-integer
  values
• a real tmp_path write of the longest allowed name

The suite passes against the production
crocodilestick/calibre-web-automated:v4.0.6 image (8/8) and on the
current main test container.

Notes for reviewers

• The 150-character limit in scripts/ingest_processor.py is left
  intact as a defensive guard for files that arrive through other paths
  (manual copy, Syncthing, NFS move). Relaxing or removing it is a
  separate audit, not part of this PR.
• The 255-byte fallback matches the POSIX NAME_MAX used by every
  supported Linux deployment. It is documented and tested for
  filesystems that reject pathconf.
• Broader Ruff cleanup of cps/editbooks.py is intentionally out of
  scope; the touched lines do not introduce new violations.

[AlfonsoDehesa]

Note on the action_required status: the upstream Test Suite workflow run completed immediately with no jobs created and no logs available on my end (GitHub purged them before I could fetch). I can't re-trigger it from my fork.

For evidence on the patch itself, the new test suite passes locally:

• Python 3.13.12 (matches actions/setup-python@v5 in .github/workflows/tests.yml)
• Run inside the official crocodilestick/calibre-web-automated:v4.0.6 image so cps imports resolve against the production runtime
• Repository's own pytest.ini and pyproject.toml

============================= test session starts ==============================
platform linux -- Python 3.13.12, pytest-9.1.1, pluggy-1.6.0
rootdir: /src
configfile: pytest.ini
plugins: anyio-4.12.0, Flask-Dance-7.1.0
collected 8 items

tests/unit/test_editbooks_ingest_path.py ........                        [100%]

============================= 8 passed in 2.30s ===============================

The broader tests/unit/ directory was not re-run because it requires Calibre CLI tools I don't have in this environment, and that broader run is outside the scope of this PR. Happy to tighten anything in the new tests or in cps/editbooks.py based on review feedback.