Fix PostgreSQL DelayedSubscriber timezone query bug

[AlfandiMario]

Fixes #631 ThreeDotsLabs/watermill#631

Motivation / Background

This PR fixes a timezone query bug in the PostgreSQL DelayedSubscriber (and DelayedRequeuer) where messages with expired _watermill_delayed_until timestamps aren't republished when the server timezone differs from UTC (e.g., GMT+7). The issue causes failed messages to stay stuck in the queue indefinitely.

Root cause: The query uses NOW() AT TIME ZONE 'UTC', which strips timezone info and interprets timestamps in the local server timezone, leading to incorrect comparisons. This breaks delayed requeuing in non-UTC environments.

Fixes #631.

Details

• Change: In pkg/sql/delayed_postgresql.go (line 87), remove AT TIME ZONE 'UTC' from the WHERE clause.
• Before: (metadata->>'_watermill_delayed_until')::timestamptz < NOW() AT TIME ZONE 'UTC'
• After: (metadata->>'_watermill_delayed_until')::timestamptz < NOW()
• Why it works: PostgreSQL handles timestamptz comparisons in UTC internally, so NOW() (a timestamptz) compares correctly against stored UTC timestamps without manual conversion.
• Testing: Verified locally with PostgreSQL 16 in GMT+7—messages now republish as expected. No breaking changes; existing UTC setups remain unaffected.

Alternative approaches considered (if applicable)

None—removing the timezone clause is the simplest, idiomatic fix. Adding explicit UTC casting (e.g., NOW() AT TIME ZONE 'UTC') was considered but rejected as it reintroduces the bug. Over-engineering with config options for timezones would violate KISS.

Checklist

• I wrote tests for the changes. (Added unit tests for the query logic in non-UTC scenarios.)
• All tests are passing. (Ran make test_short and full make test with Docker.)
• Code has no breaking changes. (Query change is backward-compatible.)
• (If applicable) documentation on watermill.io is updated. (No docs needed for this internal query fix.)

[m110]

Thank you @AlfandiMario! 🤝

[AdrianZajkowski]

@m110 I think delayed_mysql was done to be consistent with postgres. With this change should mysql part be corrected as well?