fix(sip): reject non-char-boundary Content-Length in body extraction (#108)

[ryanmurf]

What

Fixes #108 — a remotely-triggerable panic (DoS) in the SIP message parser.

parse_message() extracted the body with body[..cl] where cl is the
attacker-controlled Content-Length. Since SipMessage::parse validates the
whole datagram as UTF-8 first, body is a &str, and slicing it on a byte
index that lands inside a multi-byte UTF-8 char panics.

Impact

SipMessage::parse runs in UdpTransport::recv() which is awaited inline in
the single main SIP event loop (stack.rs:713). With the default
panic = "unwind", one malformed unauthenticated datagram unwinds and kills the
receive loop → whole-stack signaling DoS. Reachable identically over TCP/TLS/WS.

Fix

Truncate via body.get(..cl); on a non-char-boundary cl, keep the full
(already MAX_CONTENT_LENGTH-bounded) body instead of aborting. Behaviour for
well-formed messages (ASCII/boundary-aligned cl) is unchanged.

Tests

• test_content_length_mid_utf8_char_does_not_panic — body "é" (0xC3 0xA9),
  Content-Length: 1 (mid-character): asserts it parses without panic.
• test_content_length_char_boundary_truncates — body "ab",
  Content-Length: 1: asserts it still truncates to "a".

Red-control evidence (test is load-bearing)

With the fix reverted (restore body[..cl].to_string()) and the tests kept:

running 2 tests
test parser::tests::test_content_length_char_boundary_truncates ... ok
test parser::tests::test_content_length_mid_utf8_char_does_not_panic ... FAILED

---- test_content_length_mid_utf8_char_does_not_panic stdout ----
thread '...' panicked at crates/asterisk-sip/src/parser/mod.rs:799:17:
end byte index 1 is not a char boundary; it is inside 'é' (bytes 0..2 of string)

test result: FAILED. 1 passed; 1 failed

With the fix in place: 777 passed; 0 failed. cargo clippy -p asterisk-sip --all-targets is clean.

Toolchain

Rust 1.97.0. cargo test -p asterisk-sip --lib green; clippy -D warnings clean.