fix: consolidate all unmerged bug fixes from orphaned worktree branches

[rpwallin]

Summary

Consolidates unique fixes from 6 orphaned worktree branches that were each fixed independently but never merged to main — causing the same bugs to be re-fixed every session.

Bug fixes (17 fixes across 19 files)

• Portfolio: correct metric key names, safe .get() access, correlation column fix
• Chart: seaborn-v0_8 with fallback + Agg backend for headless
• Screening: days_back default 1→30 (prevents empty results)
• SQLAlchemy: wrap raw SQL with text() (3 files)
• Sentiment: low confidence returns neutral
• Health tools: handle Pydantic model_dump()
• Data: normalize timezone-aware timestamps
• Technical: missing pandas import
• Performance: SQLite guard for PostgreSQL-only queries

New features

• Finnhub provider as backup data source with fallback chain
• Agent timeout protection (25s default, depth-based for research)
• Auto-seed S&P 500 screening data on first startup
• Market overview: graceful partial failure handling

Windows compatibility

• testcontainers imports made optional
• resource module guarded (Unix-only)
• UTF-8 encoding for test file reads
• Logging handler cleanup in conftest teardown

Source branches consolidated

Branch	Unique fixes merged
dreamy-edison	Auto-seed S&P 500, matplotlib Agg backend
dreamy-mahavira	Windows test compat (testcontainers, resource, encoding)
silly-albattani	Screening days_back 1→30
eager-hertz	Timestamp normalization, sentiment, SQLAlchemy text(), Finnhub provider
thirsty-hertz	Agent timeout protection, bind_tools guard
infallible-driscoll	Correlation column fix

Test plan

• make test passes
• Portfolio endpoint works with correct metric keys
• Chart generation works without matplotlib warnings
• Screening returns results (days_back=30)
• Agent calls timeout gracefully instead of hanging

🤖 Generated with Claude Code

[sentry]

Bug: The subprocess call to scripts/seed_sp500.py uses a relative path, which will fail if the server is started from a directory other than the project root.
_{Severity: HIGH}

Suggested Fix

Construct an absolute path to the script before calling subprocess.run. Use pathlib.Path and __file__ to determine the project root and build a full, reliable path to scripts/seed_sp500.py. This will ensure the script is found regardless of the current working directory.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent.
Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not
valid.

Location: maverick_mcp/api/server.py#L1476-L1481

Potential issue: The auto-seed feature in `maverick_mcp/api/server.py` invokes a script
using `subprocess.run` with a relative path: `"scripts/seed_sp500.py"`. This path is
resolved relative to the current working directory. If the server is started from a
directory other than the project root, which is a common scenario for users invoking it
via `claude mcp add`, the subprocess will fail to find the script. The resulting
exception is caught and logged, causing the auto-seed feature to fail silently. This
leads to users seeing empty screening results, breaking a core feature of the pull
request.

[sentry]

Bug: The run_backtest_mtf method will crash with a KeyError because it accesses DataFrame columns with Title-case names ("Open") after they have been converted to lowercase ("open").
_{Severity: CRITICAL}

Suggested Fix

In the run_backtest_mtf method, change the column accessors to use lowercase names to match the data returned by get_historical_data. For example, change data["Open"] to data["open"] and weekly_data["Close"] to weekly_data["close"].

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent.
Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not
valid.

Location: maverick_mcp/backtesting/vectorbt_engine.py#L1626-L1630

Potential issue: The `run_backtest_mtf` method calls `get_historical_data`, which
returns a pandas DataFrame with column names normalized to lowercase (e.g., `open`,
`close`). However, subsequent code in `run_backtest_mtf` attempts to access these
columns using Title-case names (e.g., `data["Open"]`, `data["Close"]`). This mismatch
will cause a `KeyError` every time the function is executed, preventing the
multi-timeframe backtesting functionality from working.

[rpwallin]

Closing — merging locally instead.