feat(usace): scale kcfs flow series to cfs

[mousebrains]

Makes the usace.cda parser units-aware so the lower-Columbia dams' outflow stores correctly — a prerequisite for wiring USACE outflow onto the John Day and Bonneville gauges (companion PR #94).

What

• parse_records reads each timeseries' units field and scales kcfs → cfs in _entry_to_record. The Willamette dams (GPR/FAL) use the .Inst.0.0.Best flow series in cfs and are unchanged; the lower-Columbia dams only publish the .Ave.1Hour.1Hour series in kcfs, which would otherwise store 1000× too small.
• Aligns the code with its own class docstring ("unit conversion from kcfs to cfs") and fixes the contradictory module docstring.

Tests

kcfs→cfs scaling (pure parse_records + end-to-end on a stored observation), and cfs unchanged (GPR regression). Full suite green (991 passed).

Deploy

Deploy before PR #94, or the new USACE outflow lands 1000× small.

(Scope trimmed during review: an earlier revision also added USACE water-temperature support, dropped once we decided to wire USACE outflow only — the dam temps duplicated the gauges' existing USGS sensors.)

🤖 Generated with Claude Code

[mousebrains]

Review (updated for the scope trim) — ready to merge ✅

Re-reviewed after the simplification: the PR is now a single commit and the parser change is just the kcfs→cfs unit scaling — Temp-Water support was dropped (companion #94 now wires outflow only), so the _PARAM_MAP addition, the Temp-Water / John Day dual-key tests, and the property-test edit are all gone (4 files → 3). Good trim: smaller surface, and it drops the only speculative bit (temperature handling no consumer needed).

Verified

• Scaling logic is correct & conservative: units == "kcfs" ⇒ ×1000; cfs / ft / missing-unit pass through unchanged. Keyed off the per-timeseries units string.
• Willamette regression preserved: the .Inst.0.0.Best series reports cfs, so GPR/FAL are untouched — covered by test_parse_records_cfs_flow_not_scaled.
• Temperature now simply isn't mapped (_PARAM_MAP unchanged), so a Temp-Water series falls through data_type is None → continue and is skipped — consistent with feat(columbia): retrim lower-river gauges — USACE dam outflow, NWS stage, drop Bridgeport #94 wiring flow only.
• Tests cover kcfs→cfs (pure parse_records + an end-to-end parse→store on BON) and the cfs-unchanged regression. CI is green.

Self-contained

No fetch URLs are wired here (that's #94), so merging is a no-op in prod until #94 lands. Safe to merge first.

Minor (non-blocking)

• units = (ts_info.get("units") or "").lower() has no .strip(); a stray "kcfs " would silently skip scaling. Defensive only — the feed's unit strings are clean.

Merge order

Merge/deploy before #94.

(Supersedes my earlier comments on this PR, which were written before the temp-scope trim — and before that, corrected for a tool-output glitch. This reflects the current single-commit state.)

---

🤖 Review by Claude Code

[mousebrains]

Addressed the non-blocking nit: units now uses .strip().lower(), so a stray " kcfs " (trailing space / odd case) still scales — with a test_parse_records_kcfs_units_tolerate_whitespace_and_case regression test. Amended to keep the single-commit shape; CI re-running.