docs: drop unsupported %k from date specifier table

[prql-bot]

Problem

The date-specifier table in web/book/src/reference/stdlib/date.md is introduced as "the list of the specifiers currently supported", but %k is rejected by every dialect:

$ prqlc compile <<'EOF'
prql target:sql.postgres
from tbl
select {a = (tbl.col | date.to_text "%k")}
EOF
Error:
   ╭─[ :3:37 ]
   │
 3 │ select {a = (tbl.col | date.to_text "%k")}
   │                                     ──┬─
   │                                       ╰─── PRQL doesn't support this format specifier
───╯

chrono parses %k (and %_H) as Item::Numeric(Numeric::Hour, Pad::Space), and no dialect's translate_chrono_item in prqlc/prqlc/src/sql/dialect.rs has a Pad::Space arm — Pad::None and Pad::Zero are handled everywhere, so every space-padded specifier falls through to the unsupported-specifier error.

Solution

Docs-side fix rather than implementing space padding: no target has a direct space-padding template (Postgres would need LPAD(TO_CHAR(...), 2, ' ') rather than a format string, and MySQL's own %k is the no-pad hour, already used for Pad::None), so implementing it is a design decision rather than a doc correction.

• Drop the %k row, and note under the table that space-padded specifiers aren't supported by any dialect — so someone coming from chrono's docs isn't left guessing.
• Add a %-H row in its place. It is supported (TO_CHAR(col, 'FMHH24') on Postgres) but wasn't documented, even though the sibling %-m and %-d rows are.

Testing

Adds test_date_format_specifiers_compile to the book documentation tests, which parses the specifier table out of date.md and compiles each row against sql.postgres. This is the durable part: the table can't silently drift from the dialect implementations again. It has a floor check on the row count so it can't pass vacuously if the table's formatting changes.

Confirmed the test fails on main with exactly one error (%k) and passes with the docs fix. Full cargo test -p mdbook-prql --test documentation is green (6 passed).

Note: #6138 also edits this table (a wrong %-d example) — different row, no overlap.

---

Closes #6139 — automated triage