fix(deps): raise the scitex-dev floor to a version whose store actually works (0.20.1)

[ywatanabe1989]

The bug

0.20.0 declared scitex-dev>=0.17.8 while every store resolves through scitex_dev.store.host_store(). A consumer holding an older scitex-dev therefore resolves cleanly and then fails at use — the exact silent-failure shape the ecosystem rules forbid.

Measured on scitex-python (PR scitex-ai/scitex-python#375), which pins scitex-dev==0.28.0: uv pip install succeeded, then

[scitex-dev linter] WARNING: failed to load plugin 'clew':
    ModuleNotFoundError: No module named 'scitex_dev.store'

and TestClewIntegration::test_status, ::test_hash_file and TestCrossModuleWorkflow::test_io_save_then_clew_hash failed on all three Python legs.

The floor is measured, and importing is NOT the test

scitex_dev.store and host_store first appear in scitex-dev 0.43.1, and that version imports every symbol clew imports — a symbol-presence check would have stopped there and been wrong. A real write-then-read raises:

TypeError: tuple indices must be integers or slices, not str
  in scitex_dev/store/_codec.py:108  row_from_db

So the probe is an actual record-and-read-back against PostgreSQL, using the published 0.20.0 wheel from PyPI in a fresh venv per version:

scitex-dev	result
0.28.0	FAIL — no scitex_dev.store at all
0.43.1	FAIL — imports fine, rows() TypeErrors
0.46.0	FAIL — same
0.48.0	FAIL — same
0.49.0	FAIL — same
0.49.2	OK
0.49.3	OK
0.50.0	OK
0.57.0	OK

The change

scitex-dev>=0.49.2 in both the runtime dependency set and dev, so the bad combination is refused at resolve time instead of failing later. Version 0.20.1. No code change — this is 0.20.0 with honest metadata.