fix: show microsecond precision in borg diff when timestamps differ only at sub-second level (fixes #9147)

[hiepau1231]

Summary

Fixes #9147 — borg diff on FreeBSD/NetBSD shows confusing [ctime: Wed, 2025-11-05 17:45:53 +0000 -> Wed, 2025-11-05 17:45:53 +0000] output where both timestamps look identical but actually differ at nanosecond level.

• Add format_timestamp_pair(ts1, ts2) in helpers/time.py: selects second-precision format normally, but upgrades to microsecond precision (%H:%M:%S.%f) when two timestamps appear equal at second resolution but differ at sub-second level
• Update DiffFormatter.format_time() in helpers/parseformat.py to use format_timestamp_pair(), so diff output shows [ctime: ...17:45:53.000123 ... -> ...17:45:53.000456 ...] instead of the confusing identical-looking pair
• Re-enable test_hard_link_deletion_and_replacement on FreeBSD, NetBSD, and OpenBSD (was previously skipped): the test now accepts a POSIX-valid sub-second ctime update on surviving hardlinks while verifying no content/mode/mtime changes occurred

Root cause

On BSD, creating a new file at a previously hard-linked path causes a sub-second ctime update on surviving hard links (POSIX-valid behaviour). The old format_time() always used second-precision format, so two timestamps like 17:45:53.000123 and 17:45:53.000456 both rendered as 17:45:53, producing the confusing X -> X output.

Test plan

• test_format_timestamp_pair_different_seconds — second-precision when timestamps differ by ≥1s
• test_format_timestamp_pair_same_second_different_microsecond — microsecond-precision for sub-second diff
• test_format_timestamp_pair_identical — second-precision for identical timestamps
• test_diff_formatter_format_time_shows_microseconds_when_same_second — full formatter path
• test_diff_formatter_format_time_no_microseconds_for_different_seconds — control case
• test_hard_link_deletion_and_replacement on FreeBSD/NetBSD/OpenBSD CI — required before merge

⚠️ The BSD-specific fix can only be validated on FreeBSD/NetBSD/OpenBSD CI runners. All Linux tests pass locally.

[ThomasWaldmann]

Thanks for the PR!

That variable formatting is a mixed blessing:

• for human readers, it might be the best way
• for automated parsing, it is the worst if the format is not constant

Can you fix the linting issues, so the BSD tests are not cancelled?

[hiepau1231]

Thanks for the review!

Regarding variable formatting: noted — the trade-off is real. The current implementation only switches to microsecond precision in the narrow case where two timestamps are identical at second resolution but differ sub-second, so the output format stays constant for the vast majority of diffs. That said, I'm open to alternative approaches (e.g. always using microsecond precision, or a flag) if you prefer a guaranteed constant format.

Fixed the black formatting issues in diff_cmd_test.py and parseformat_test.py — the @pytest.mark.skipif reason arg and ItemDiff() calls have been collapsed to single lines as black requires. CI lint should pass now.