event_rabbitmq: fix dupl_string() NUL-inclusive len corrupting AMQP shortstr

[NormB]

Summary

dupl_string() in event_rabbitmq.c increments dst->len after NUL-terminating the unescaped string (dst->len++), causing .len to include the trailing NUL byte. This makes amqp_basic_publish() encode exchange and routing-key as AMQP shortstr fields with an extra 0x00 byte, breaking broker routing.

The bug was introduced in commit 0260de0c1 (2014) when dupl_string() was refactored to use un_escape(). The original memcpy-based implementation had the same off-by-one (dst->len = end - begin + 1).

Details

AMQP 0-9-1 encodes exchange and routing-key as shortstr fields (1-byte length prefix + N content bytes). With the inflated .len, the broker receives a shortstr containing the field value followed by 0x00, which fails to match the declared exchange or creates one with an embedded NUL.

Over time, several - 1 compensations were added downstream to work around the inflated length:

• tls_dom_name.len-- in rmq_parse() (commit 472ae546c, TLS support)
• address.len - 1, exchange.len - 1, routing_key.len - 1 in rmq_print()
• strlen(user) - 1 in rmq_print() — this one was wrong even pre-patch since strlen() never counts the NUL

Solution

Remove the dst->len++ in dupl_string() and all downstream compensations. Three additional issues fixed while testing:

1. Memory leak: dupl_string() calls shm_malloc before un_escape(), but on un_escape() failure the buffer was never freed. Added shm_free + cleanup.
2. Default user NUL termination: shm_malloc(rmq_static_holder.len) + memcpy(..., len) for "guest" didn't copy the NUL terminator. The pointer is passed to amqp_login() → strlen(). Changed to len + 1.
3. strlen(user) - 1: The /* skip 0 */ comment reveals the author believed strlen() includes the NUL. It doesn't — this always truncated the last character of the username in rmq_print() output.

Tested with:

• AMQP wire capture (tshark) confirming correct shortstr encoding
• TLS domain name lookup (instrumented debug + regression simulation)
• Standalone test harness (1019 lines) with 100K-iteration stress test confirming shm leak fix

Compatibility

No backward compatibility issues. All behavioral changes are strictly correctional:

• Wire encoding: NUL byte no longer included (fixes the reported bug)
• rmq_match(): symmetric — both sides were inflated equally, no change
• rmq_print(): compensations removed, output unchanged
• amqp_login() user/password: uses char* via strlen(), unaffected

Closing issues

Closes #3828