Fix all compiler warnings in the ESP-IDF build

[eman]

Summary

The component emitted 21 compiler warnings in the ESP-IDF build; this brings it to zero. All 21 were format-string mismatches — no logic changes.

-Wformat= (19)

On ESP32, uint32_t is long unsigned int, so %u/%d didn't match the arguments being logged: millisecond timers, event/schedule timestamps, start counts, the service-discovery retry delay, and the BLE passkey. These now use PRIu32 from <cinttypes>.

ESPTime::timestamp is printed via a long long cast rather than a fixed specifier — time_t is long long on ESP-IDF but long on many hosts, and these sources are also compiled on the host by the test suite, so one cast keeps a single format string correct on both.

-Wformat-truncation= (2)

The get_begin_time() / get_end_time() helpers in schedule_entry.h format "HH:MM" into a char[6]. The compiler can't prove a uint8_t hour/minute renders in two digits, so it flagged possible truncation. Both fields are now bounded to two digits, which silences the warning and also keeps output well-formed if the pump ever reports an out-of-range byte.

Files touched

alpha_hwr.cpp, ble_connection_manager.cpp, device_info_service.cpp, event_log_service.cpp, schedule_entry.h, schedule_service.cpp, time_service.cpp, transport.cpp

Testing

• Full ESP-IDF build (ESPHome 2026.7.0, esp32-c3): 0 warnings, down from 21. Verified every previously-warning translation unit was actually recompiled, not served from cache.
• Host suite (cd tests && make clean && make test, built with -Wall -Wextra): 0 warnings, 494 assertions passing, 0 failures.
• Not yet flashed to hardware. Per AGENTS.md §4 that's the normal gate, but this change only alters format specifiers and the values handed to them, so runtime behaviour is unchanged apart from log lines now rendering correctly instead of relying on ESP32 ABI luck.