BulkCopy sub-second timeout truncates to zero and becomes infinite

[saurabh500]

Describe the bug

BulkCopy::timeout(Duration) converts the supplied duration to whole seconds using timeout.as_secs() as u32 at mssql-tds/src/connection/bulk_copy.rs:435. Any positive duration below one second therefore becomes 0.

The bulk-copy timeout path defines 0 as an infinite timeout (BulkCopyTimeoutState::from_seconds(0) has no deadline), so requesting a short timeout silently disables timeout enforcement.

Steps to reproduce

1. Configure a bulk copy with a positive sub-second timeout:

bulk_copy.timeout(Duration::from_millis(500));

2. BulkCopy::timeout stores timeout_sec = 0 because Duration::as_secs() truncates fractional seconds.
3. Start the bulk-copy operation.
4. Observe that the timeout state treats 0 as infinite and never expires.

The conversion can also be demonstrated directly:

assert_eq!(Duration::from_millis(500).as_secs() as u32, 0);
assert!(BulkCopyTimeoutState::from_seconds(0)
    .remaining_duration()
    .is_none());

Expected behavior

A positive timeout must never become an unlimited timeout. Since the current public API accepts Duration, sub-second values should either retain their precision or be converted safely without producing the 0 = infinite sentinel (for example, round a positive fractional duration up to one second).

Actual behavior

Every timeout from 1 ms through 999 ms is truncated to 0 and interpreted as unlimited. A caller requesting the shortest timeout receives no timeout at all.

Version

main at 068efe7a

Affected crate

mssql-tds

Environment

All platforms; the defect is in platform-independent duration conversion.

Additional context

Related to #273, which tracks the broader class of arithmetic producing the 0 = unlimited sentinel. This issue isolates the directly reachable public BulkCopy::timeout(Duration) case so it can be fixed and regression-tested independently.

PR #270 does not address this path.