fix(nwis): name the peak date format instead of making pandas guess

[thodson-usgs]

Why

preformat_peaks_response calls pd.to_datetime(df["peak_dt"], errors="coerce").
On a column mixing zero-filled and parseable dates pandas cannot find one format
that fits both, falls back to per-element dateutil parsing, and says so:

UserWarning: Could not infer format, so each element will be parsed
individually, falling back to `dateutil`.

That mix was rare until #395, which stopped dropping peaks whose date NWIS only
partly knows. Keeping them made the mix the normal case for any long
historical record — so the warning now fires on ordinary get_discharge_peaks
calls. Library noise on stderr, for behaviour the library intends.

Found while reviewing #395 against the glossary; it is a separate concern from
that review, so it is a separate PR.

The fix

The format was never actually ambiguous. The RDB header types peak_dt as
10d — a ten-character date — and any time is carried separately in peak_tm:

agency_cd  site_no   peak_dt  peak_tm  peak_va  ...
5s         15s       10d      6s       8s

So %Y-%m-%d is the only shape the column takes, and naming it stops pandas
guessing.

Verified identical, not assumed

Against the real tests/data/waterservices_peaks.txt column plus zero-filled,
blank, malformed and time-bearing values:

input shape	warns today	warns after	results identical
all YYYY-MM-DD	no	no	✅
zero-filled + parseable	yes	no	✅
all zero-filled	yes	no	✅
carrying a time component	no	no	✅
malformed	no	no	✅

The warning fires only on the case #395 made normal. A value carrying a time
coerces to NaT either way, so nothing that parsed before stops parsing.

The test can fail

test_preformat_peaks_response_is_quiet_on_partly_dated_peaks asserts silence
with warnings.simplefilter("error"). It was run against the unfixed
function first and confirmed to fail there:

FAILED tests/nwis_test.py::test_preformat_peaks_response_is_quiet_on_partly_dated_peaks
  UserWarning: Could not infer format ...

1140 tests pass, mypy --strict clean, all hooks green.

Scope

Independent of #400 and #402. Touches nwis.py, one new test, and a NEWS entry.

🤖 Generated with Claude Code

https://claude.ai/code/session_01JAEQqs7XzQHGQQi2KakuXD