Cleaned up the protocol layer and metadata

- request() now reads exactly the expected response length with
  read_exact instead of a single non-guaranteed fill_buf, fixing a
  latent bug where a response split across multiple TCP packets could
  be mistaken for malformed data; the per-method length checks are now
  redundant and were removed
- sequence_number wraps explicitly via wrapping_add instead of
  relying on implicit (and debug-vs-release inconsistent) overflow
  behavior after 65536 requests
- get_name validates UTF-8 strictly instead of silently lossy-decoding
  invalid bytes; on invalid data it now returns the FromUtf8Error
  itself (byte-level \0 trimming moved after the UTF-8 check), letting
  the caller choose to fail or fall back to a lossy rendering
- removed the unused Default derive from Clock and Record
- removed npm from the CI image (no JS/Node in this repo), fixed a
  typo in a CI step name
- fixed the Cargo.toml license field to the actual SPDX identifier
  (AGPL-3.0-or-later, per the notice at the end of LICENSE) instead of
  the free-text "GNU AGPLv3.0"

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

README.md

@@ -45,11 +45,11 @@ If you want to read about this reverse engineering attempt, check out
        println!("No\tMchn\tEnNo\t\tName\t\tMode\tIOMd\tDateTime\t");
        while let Some(record) = record_stream.next().await {
            let record = record.unwrap();
-           let name = r701
-               .get_name(record.employee_id)
-               .await
-               .unwrap()
-               .unwrap_or(format!("user #{}", record.employee_id));
+           let name = match r701.get_name(record.employee_id).await.unwrap() {
+               Some(Ok(name)) => name,
+               Some(Err(error)) => String::from_utf8_lossy(error.as_bytes()).into_owned(),
+               None => format!("user #{}", record.employee_id),
+           };
 
            println!(
                "{:0>6}\t{}\t{:0>9}\t{: <10}\t{}\t{}\t{}",