Reset parser after inter-byte timeout to prevent truncated frame corruption

[brocci]

As reported in issue #28:

[BUG] No inter-byte timeout: a truncated frame corrupts the next frame and applies garbage to outputs

• Status: still present in v1.7.0 (unchanged; INIT now also flows through this path).
• Spec: LCS-9.10.1 p.2 (valid-message definition), p.8 §D.2 (the host has a timeout concept; nodes get no guidance — see ambiguities).
• Code: src/CMRI.cpp:214-233 (DECODE_DATA, DECODE_ESC_DATA); no timer anywhere in the library.
• Two concrete cases, both verified against the state machine:
  1. A T frame that dies before its ETX leaves the parser in DECODE_DATA. The next real frame's FF FF 02 UA MT bytes are stored into _rx_buffer as if they were output data (until the buffer fills), and the next frame's ETX terminates the ghost frame and returns SET — the sketch then drives layout outputs with 0xFF 0xFF 0x02... garbage.
  2. A frame that dies immediately after a DLE leaves the parser in DECODE_ESC_DATA, which unconditionally consumes the next byte (src/CMRI.cpp:226-233) — i.e., the first SYN of the next frame is eaten as "escaped data," cascading into case 1.
• Failure scenario: one glitched/brownout-truncated transmission produces visibly wrong turnout/signal states on the next host update, not just a lost frame.
• Fix: reset _mode to PREAMBLE_1 (and discard the partial buffer) if more than a few character times elapse between bytes; or double-buffer and commit only on ETX.

This PR implements the required fix and adds corresponding tests. This has not been tested on real hardware.

The default INTER_BYTE_TIMEOUT_MS has been calculated based on serial transmission at 9600 bauds 8N2. JMRI uses 19200 bauds as default baud rate for C/MRI, but this default timeout value matches the baud rate used in all examples of this library.