What does this implement/fix?

Fix millis overflow handling in SprinklerValveOperator using wrapping 32-bit subtraction instead of absolute deadline comparisons. Also fix several pre-existing bugs found during review.

Approach

Rather than switching to 64-bit millis_64() (which has performance overhead in the hot loop() path — called ~7100 times/min), use the standard embedded pattern of wrapping subtraction: (now - start) > delay instead of now > (start + delay). This correctly handles the 49.7-day millis() rollover with zero additional cost, using the already-cached App.get_loop_component_start_time().

The start_millis_ and stop_millis_ fields are changed from uint64_t to optional<uint32_t> to eliminate the use of 0 as a sentinel value (which is a valid millis() timestamp that occurs every 49.7 days).

Bugs fixed:

loop() — valve stops mid-run after millis wrap:
The code compared uint32_t now against uint64_t start_millis_ deadline. After wrap, now (promoted to 64-bit) could never reach the 64-bit deadline, so it fell into the else branch which called stop() with a TODO comment: "perhaps millis() rolled over...or something else is horribly wrong!". With wrapping subtraction, the comparisons work correctly and the bailout is removed.

time_remaining() — returns ~49.7 days remaining after wrap:
The deadline start_millis_ + delay + duration exceeds UINT32_MAX but millis() can never reach it as a 32-bit value, so the valve appears stuck running forever. Now uses elapsed-time subtraction which handles wrap correctly.

time_remaining() — unsigned underflow when stopped during start delay:
Pre-existing bug: if a valve was stopped during the start_delay_ phase (before the run began), run_duration_ - (stop - start) would underflow since the elapsed time could exceed run_duration_. Now correctly detects this case and returns the full run duration.

time_remaining() — millis zero sentinel unreliable:
Pre-existing bug: start_millis_ == 0 was used to mean "not started", but 0 is a valid millis() value. Changed to optional<uint32_t> with .has_value() checks, matching the pattern used in [pn532] Replace millis zero sentinel with optional (#14295).

total_cycle_time_enabled_incomplete_valves() — unsigned underflow:
incomplete_valve_count-- wraps to UINT32_MAX when both incomplete and enabled counts are zero, producing a bogus cycle time.

set_repeat() — unchecked optional dereference:
Pre-existing bug: repeat.value() called without has_value() check. Changed to value_or(0) to match the usage two lines below in the same function.

queue_valve() — unchecked optional dereference:
Pre-existing bug: run_duration.value() called without has_value() check. Changed to value_or(0) to match the log statement in the same function.

Types of changes

• Bugfix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)
• Developer breaking change (an API change that could break external components)
• Code quality improvements to existing code or addition of tests
• Other

Related issue or feature (if applicable):

N/A

Pull request in esphome-docs with documentation (if applicable):

N/A

Test Environment

• ESP32
• ESP32 IDF
• ESP8266
• RP2040
• BK72xx
• RTL87xx
• LN882x
• nRF52840

Example entry for config.yaml:

# No config changes

Checklist:

• The code change is tested and works locally.
• Tests have been added to verify that the new code works (under tests/ folder).