java.net host identity, and a real java.util.Properties

[yogthos]

Closes jolt-6227.

The bead was filed from clj-uuid's comments; reading its node.clj showed those comments describe the approach it rejects. make-node-id never calls .getHardwareAddress — it digests the machine's identity strings instead. So the gap is different from what was filed, and this implements what the code actually reaches for.

java.net

InetAddress/getLocalHost, getAllByName (an array, so alength/aget hold), .getCanonicalHostName, .getAddress, .isLoopbackAddress, .equals/.hashCode; and a new java.net.NetworkInterface — getNetworkInterfaces, getByName, getByInetAddress, .getInetAddresses, .getHardwareAddress, .getDisplayName, .isLoopback — over getifaddrs(3), gethostname(2) and getnameinfo(3). IPv4, matching the rest of jolt.socket.

getifaddrs reports one entry per address, so an interface with an address and a MAC appears twice; the entries are grouped by name into one interface each, which is the shape java.net presents. An address read off an interface carries no hostname — the JVM prints it /127.0.0.1 — and .getHostName resolves once and caches, rather than putting a DNS round trip per address in the way of enumerating interfaces.

Touching any java.net class now loads jolt.socket on demand, the courtesy java.time and MessageDigest already had. A JVM library reaching for InetAddress should no more have to know which namespace installs it than it does on the JVM.

.getHardwareAddress and getByInetAddress are here even though clj-uuid doesn't use them: they're the documented java.net surface, and the getifaddrs walk that answers the rest already has the data.

java.util.Properties

Found along the way, and needed for the same caller: System/getProperties returned a plain map, so .getProperty read as a key lookup and answered nil for every system property. There's now a real Properties with the JVM's defaults chain — and its exact span: getProperty, propertyNames and stringPropertyNames see the defaults; the inherited Hashtable surface (get, containsKey, keySet, values, size) does not. A Properties holding only defaults answers isEmpty true with an empty keySet while propertyNames lists them.

System/getProperties holds its values as real entries, as on the JVM, rebuilt per call so user.dir and java.class.path stay current instead of freezing at the first call, and writes through to the store System/getProperty reads.

While in there, the three spellings of "is this a hashmap-backed host object" were one predicate's worth of drift; they are one now, so a shim of that shape cannot be countable through one accessor and opaque through another.

Validation

• make test green: exit 0, 82 CI targets, 0 new divergences, 0 failures
• 30 new checks in the jolt.socket gate (make smoke greps its marker), each expectation certified against JVM Clojure 1.12.5 first — including the defaults-span split and the lazy-resolve-and-cache behaviour
• 4 cold-process smoke cases for the autoload path (a fresh process touching java.net with no require)
• InetAddress/getLocalHost renders byte-identically to the JVM on this machine
• clj-uuid's all-local-addresses + make-node-id run end to end and are stable across calls; the config library's (->> (System/getProperties) (map (fn [[k v]] …))) still resolves
• Site docs updated (host-interop.md) — the "gated behind (require 'jolt.socket)" note is no longer true