fix(ld2451): fix FrameParser return semantics, skip config frames, document direction byte

[poolski]

Summary

• Fix FrameParser::pop() return value on tail mismatch — previously returned true (implying a frame was produced) when the tail bytes didn't match and only 1 byte was advanced. Now returns false, consistent with the API contract that true means a complete frame was consumed.
• Fix extract_frame_() frame_produced tracking — the loop() drain loop was setting parsed_frame = true even for sync-error retries (tail mismatch advances), silently suppressing the RX-activity diagnostic log. Refactored to separate "should keep draining" (return value) from "did a real frame get parsed" (new bool &frame_produced reference param).
• Skip config/ACK frames cleanly in both parsers — the sensor emits two frame types: data (F4 F3 F2 F1 header) and config/ACK (FD FC FB FA header). Previously a config ACK would be consumed byte-by-byte as the parser searched for the data header, causing spurious log noise and delaying the next real frame. Both FrameParser::pop() and LD2451Component::extract_frame_() now detect the config header, read the 2-byte LE length, and erase the complete config frame in one operation.
• Document direction byte convention — cross-referenced three external implementations (jbeale monitor, Fiooodooor LD245X, Madproforg HLK_LD2451) to resolve an ambiguity in the official PDF. Added a comment in types.h confirming 0x00 = Approaching, 0x01 = Moving away.

Test plan

• New tests in frame_parser_test.cpp: config frame followed by data frame parses correctly; tail mismatch returns false
• All existing host tests still pass (bash components/ld2451/tests_host/run_host_tests.sh)
• Smoke test on hardware: verify no regression in normal target detection and that config-frame log noise is gone at startup

Note

I used Claude Code for some of these changes