Fix Python DTLS PSK handshake compatibility and probe validation

[KRZ303]

This follows the Python DTLS proposal in #115.
I tested the corrected probe against my oven inside HA Core on October 3.
It authenticated with the complete 16-byte PSK identity containing a zero byte.
One GET /oic/d returned CoAP 2.05 and my oven's expected device ID.

What needed fixing

• Keep the initial ClientHello in the transcript when the server skips the cookie exchange.
• Accept DTLS 1.0 framing for the initial plaintext HelloVerifyRequest, while retaining DTLS 1.2 checks elsewhere.
• Use unconnected UDP so replies from another source port can arrive.
• Allow a fixed local port so the probe can reuse the integration's existing peer endpoint.
• Validate response correlation, CoAP 2.05, the CBOR map, and the expected device ID before reporting success.
• Handle empty ACKs and acknowledge separate CON responses.
• Attempt shutdown and socket closure after success or failure.
• Remove addresses and identity/key fingerprints from output.
• Document the probe's required runtime dependencies: cryptography and cbor2.

Result on my oven

identity      : 16 bytes, contains a zero byte: yes
handshake     : OK in 0.1s (DTLS 1.2, ECDHE-PSK-AES128-CBC-SHA256)
GET /oic/d    : OK (2.05, CBOR map, expected device ID matched)
close_notify  : sent

The probe exited with code zero and no stderr.
I temporarily unloaded only the oven entry and restored it afterward.
The oven resumed Observe mode, and both oven and washer entries were loaded.
close_notify: sent records local UDP submission; I did not measure its receipt on the oven.

Checks

• All 26 regression tests passed, including real OpenSSL exchanges with and without cookies.
• Negative tests cover malformed empty CoAP messages, rejected non-HVR DTLS 1.0 records, and local-port bind failure with cleanup.
• A separate local Mbed TLS 3.6.7 fixture passed both exchanges with encrypt-then-MAC and extended master secret disabled.
• The README records the tested dependencies and the command for reproducing the 26-test suite.

Since the hardware run, the probe gained dependency instructions in its module docstring; its executable code and the engine are unchanged.
The additional regression tests ran offline. I did not repeat the hardware test.

This PR updates the standalone scripts on scratch/test-scripts.
Integration into the library remains separate.
The installed HA integration and protocol library were unchanged by this probe.