impute_time() truncates sub-hour UTC offsets: tz = "+0530" is recorded as +05:00

`impute_time.partial_time()` assigns the imputed offset with integer division:

```r
# R/impute.R
tzhour_na <- is.na(vctrs::field(impute_pttm, "pttm_mat")[, "tzhour"])
vctrs::field(impute_pttm, "pttm_mat")[tzhour_na, "tzhour"] <- tz %/% 60
```

`interpret_tz()` returns minutes, so `%/% 60` discards any part of the offset that is not a whole hour. `tzhour` is stored as a double and holds fractional offsets perfectly well elsewhere in the package, so the loss is entirely in this line.

### Reproducible

```r
library(parttime)
options(parttime.assume_tz_offset = NA) # so the offset is genuinely unknown
x <- as.parttime("2001-06-15")

impute_time_mid(x, tz = "+0530") # tzhour 5 should be 5.5
impute_time_mid(x, tz = "+0545") # tzhour 5 should be 5.75
impute_time_mid(x, tz = "-0430") # tzhour -5 should be -4.5
impute_time_mid(x, tz = "+1245") # tzhour 12 should be 12.75
impute_time_mid(x, tz = "-0400") # tzhour -4 correct, whole hour
```

This affects explicit offset strings, not only timezone names, so it is independent of how the offset was obtained.

### Effect

`%/%` is floor division, so the recorded offset is always rounded *west*, by up to 59 minutes:

| offset | minutes | `%/% 60` | correct | error |
|---|---|---|---|---|
| `+0530` | 330 | 5 | 5.5 | −0:30 |
| `+0545` | 345 | 5 | 5.75 | −0:45 |
| `-0430` | −270 | −5 | −4.5 | −0:30 |
| `-0545` | −345 | −6 | −5.75 | −0:15 |

A local time carrying an offset that is too far west resolves to a UTC instant that is correspondingly *late*, so `10:30 +05:30` — which is `05:00Z` — is recorded as `10:30 +05:00` and resolves to `05:30Z`.

Every sub-hour zone is affected: India and Sri Lanka (+05:30), Nepal (+05:45), Iran (+03:30), Afghanistan (+04:30), Myanmar (+06:30), Eucla (+08:45), Adelaide and Darwin (+09:30), Lord Howe (+10:30), Chatham (+12:45), Newfoundland (−03:30), Marquesas (−09:30).

The package's own defaults are all whole hours — `impute_time_min()` uses `"-1200"`, `impute_time_max()` uses `"+1400"`, `impute_time_mid()` uses `"GMT"` — so the defaults are unaffected and this only appears when a caller supplies a sub-hour `tz`.

### Fix

```r
vctrs::field(impute_pttm, "pttm_mat")[tzhour_na, "tzhour"] <- tz / 60
```

`tzhour` already carries fractional values elsewhere — casting a `POSIXct` in `Asia/Kathmandu` yields `tzhour = 5.75`, and `format()` renders it as `+05:45` — so nothing downstream needs to change to accommodate it.

I could not find any test that pins the current truncated values, so the change should not require test updates beyond adding coverage for it.

### Not affected

For completeness, two nearby things that look similar but are correct:

- `to_gmt()` uses `%/% 1` on `tzhour` for the hour, but carries the remainder into minutes on the following line (`x[, "min"] + x[, "tzhour"] %% 1 * 60`), so the fraction survives.
- `format()` pairs `%/% 1` with `%% 1 * 60` the same way, and renders `+05:45` and `+05:30` correctly.

One genuinely odd spot, though it appears unreachable: `impute_partial_time_to_chr()` formats the offset minutes as `tzhour %/% 1 * 60`, which yields `300` rather than `30` for a `+05:30` offset (and `300` for a whole-hour `+05:00` too). Nothing in the package appears to call that function, so it may just be dead code worth removing.

Issue created with the help of Claude Opus 5.

Start in R/impute.R at impute_time.partial_time(), where the missing tzhour value is assigned from interpret_tz(). Reproduce the issue with the documented impute_time_mid() calls for +0530, +0545, and -0430, then add coverage showing fractional offsets are preserved while whole-hour offsets remain unchanged. Done means the recorded tzhour values match the supplied offsets and the relevant package tests pass.

Written by the indexing model from the issue text.