fix(data): short-history branch matches stored dtype per-column

[cipher813]

Summary

Follow-up to PR #76. The short-history write path hardcoded `Volume → int64`; stored ArcticDB schema varies per-ticker (some int64, some float64 depending on backfill vintage). Fix: astype every column to `hist.dtypes[col]`, authoritative by construction.

Why

PR #76 shipped + validated today — today's EOD email sent successfully because SNDK's stored Volume was already int64 (matching the hardcode). Three other short-history tickers (SOLS, ULS, +1) failed the update with `FLOAT64/INT64` mismatch and contributed n_err=4 in the daily_append summary. Tomorrow's unattended 13:05 PT daily_append would repeat the same failures every day until fixed.

Change

```python

before (PR #76)

for col in new_row.columns:
if col in OHLCV_COLS:
if col == "Volume":
new_row[col] = new_row[col].astype("int64")
else:
new_row[col] = new_row[col].astype("float64")
else:
new_row[col] = new_row[col].astype("float32")

after

for col in new_row.columns:
new_row[col] = new_row[col].astype(hist.dtypes[col])
```

Test plan

• `pytest tests/test_daily_append_semantics.py -v` — 8/8 pass (new dtype test + 7 prior)
• Full repo suite: `pytest tests/` — 111/111 pass
• Tomorrow's unattended 13:05 PT daily_append: expect n_partial count to increase + n_err to drop by 3 (SOLS, ULS, +1 previously failing)

🤖 Generated with Claude Code