Treat CLI semantic failures as errors, clarify graph materialization, add tests and docs

[sean2077]

Motivation

• Prevent silent false-successes when the CLI exits with code 0 but returns JSON { "ok": false } so automation relying on check=True raises on semantic failures.
• Reduce terminology drift by aligning package messaging to the preferred "Python preview client" phrase and clarify how in-memory graphs are materialized with regard to filename suffixes.
• Add explicit unit tests that lock in intended exception vs return behavior for the semantic-failure case.

Description

• Normalize package docstrings to use the term Python preview client and adjust a few related docstrings for clarity (python/topoexec_preview/__init__.py and python/topoexec_preview/client.py).
• Change _json_command in TopoExecClient to treat check=True as failing when either returncode != 0 or the parsed JSON contains "ok": false, by computing semantic_ok and raising TopoExecCommandError accordingly.
• Add a clarifying docstring to materialized_graph that documents the suffix-preserving behavior for in-memory GraphDocument inputs and defaulting to .yaml only when no suffix is present.
• Add a development doc file docs/20-development-overview/issue-task-proposals.md listing related follow-up tasks.
• Add unit tests in tests/python/test_python_preview.py which mock subprocess.run to return stdout='{"ok": false, ...}' and assert that check=True raises TopoExecCommandError while check=False returns a CommandResult with ok == False.

Testing

• Ran the new unit tests in tests/python/test_python_preview.py via python -m unittest tests/python/test_python_preview.py and the added tests passed.
• Existing smoke-style tests in the same file were left unchanged and were not modified by this PR's logic, and the test suite showed no regressions in local runs.

---

Codex Task