Fix #289: resolve 11 pre-existing test failures on main

[ericleepi314]

Closes #289

Summary

Full pytest tests on main was red with 13 failures; 2 are tracked separately in #274. This fixes the other 11, each at its root cause:

Group	Root cause	Fix
providers (5 tests)	Implementation intentionally wraps SDK clients with with_options(timeout=…, max_retries=…) (stalled-stream protection); test mocks predated it	Mock with_options as pass-through + new TestApplyClientTimeout (3 tests) covering the wrap kwargs, env overrides, and the failure fallback
parity defaults (2 tests)	TS-default check applied to ALL registered tools, but advisor/Workflow are Python-only and legitimately read-only/concurrency-safe	Scope the default checks to TS core tools (from ts_tool_names.json)
MCP validation (2 tests)	command="python" absent from PATH on python3-only systems	Use sys.executable
web_fetch (1 test)	Regex fallback _NOISE_BLOCK_RE missed iframe|object|embed which the bs4 path strips — real implementation bug, iframe content leaked into extracted text	Add them to the alternation
bridge (1 test)	Untracked stale src/services/bridge/__pycache__ in a checkout survives branch switches and imports as a namespace package	Use find_spec; tolerate origin is None (namespace pkg), still fail if a real shim (__init__.py) returns

Also adds testpaths = ["tests"] per the issue's suggestion so bare pytest no longer aborts collection on reference_projects/.

Test plan

• All 11 originally-failing tests now pass
• Full suite: 2 failed, 7760 passed, 5 skipped — the 2 are exactly the ToolContext.permission_context defaults to bypassPermissions, silently disabling the workspace-root allowlist #274-tracked tests (out of scope per issue text)
• Bridge fix verified by simulating the stale __pycache__ dir
• New timeout tests verified immune to ambient CLAWCODEX_LLM_* env vars
• Critic review loop: APPROVE after 2 revision rounds

🤖 Generated with Claude Code