test: memory status B1 regression + sync_helpers coverage (12 new tests)

[FZ2000]

Coverage for the two gaps identified in the Forge audit: memory path in _tool_sync_status (B1 regression) and zero coverage on sync_helpers.

New: tests/test_manifest.py — 3 memory status tests

Added to TestToolSyncStatus:

Test	What it catches
test_synced_when_memory_file_present	The B1 crash — old code called .get() on strings from .values(). This would AttributeError before the fix in PR #18.
test_out_of_sync_when_memory_file_deleted	Memory file recorded in manifest, then deleted → "out of sync"
test_synced_when_memory_not_yet_recorded	Manifest has only MCP servers, memory dict is {} → no false "out of sync"

The memory tests failed against main (confirming they catch the real bug) and pass after cherry-picking the fix from fix/bugs.

New: tests/test_sync_helpers.py — 12 tests

TestResolveTargetTools (5 tests)

• --tools flag parsing + whitespace strip
• Empty string returns []
• --all calls detect_installed_tools
• --all with no detected tools returns []

TestSyncAll (6 tests)

• Happy path returns True
• All three phases called per tool (apply_skills, apply_mcp_servers, apply_memory_via_llm)
• --no-memory skips apply_memory_via_llm
• Partial failure (1 of 2 tools errors) → True (any_success)
• All fail → False
• Single tool success

TestSyncSkillsPerToolCounter (1 test)

• Verifies per-tool success message shows 3 copied not 6 copied for 2 tools × 3 skills. Catches the B2 counter bug directly.

Depends on

Cherry-picks fix/bugs (e49818b) so the memory tests pass.
Intended to be merged after PR #18 lands on main.

Test count

137 passing (122 from PR #18 base + 15 new).