ISOMsg: strict path readers, dataset element writers, dataset-aware clone

[ar]

Second PR for #790, following the implementation plan. PR #791 was the first. Closes #790.

New API (ISOMsg)

• Optional<Object> findValue(String fpath) throws ISOException
• List<DatasetElement> findDatasetElements(String fpath) throws ISOException
• setDatasetElement(fpath, elementId, value) / setDatasetElement(fpath, datasetId, elementId, value), plus withDatasetElement and unsetDatasetElement counterparts
• DatasetElement.isConstructedTag(int), replacing private copies in DatasetPackager and XMLPackager (separate commit)

Behavior

Path grammar (new methods and clone(String...) only; the existing set/unset/hasField/getValue/getString/getComponent(String) stay lenient): segments are [0-9]+ or 0x[0-9a-fA-F]+, in the range 0..Integer.MAX_VALUE. Empty segments, signs, whitespace, 0X and out-of-range values throw IllegalArgumentException naming the path and the bad segment, whatever the message contains.

Readers

• Every segment but the last must be a sub-ISOMsg. An absent field, or a path through a leaf, returns empty.
• Dataset descent: if the field packager of the message holding the dataset field is a DatasetFieldPackager, hasDatasetEnvelope() decides the layout, and a mismatched path returns empty. Otherwise the layout is inferred from the remaining segments (1 = no envelope, 2 = envelope, more = empty). Nested dataset fields use the sub-message's packager.
• findDatasetElements collects matches from every dataset with the addressed id. findValue returns the first, so findValue(p).isPresent() ⇔ !findDatasetElements(p).isEmpty().
• Constructed elements are returned as encoded byte[]; the Javadoc points to TLVList for decoding children.

Writers

• setDatasetElement needs no packager and creates intermediate composites. Ints carry the dataset coordinates, because a single string path is ambiguous when there's nothing to walk yet.
• New datasets: no envelope → TLV; envelope → TLV up to 0x70, DBM above. Elements are replaced (putElement). TLV elements record the constructed flag. A null value removes the element.
• A leaf in the way, or a non-dataset field at the target, throws ISOException. The value type is checked before anything is created.

clone(String...): copies every occurrence of a dataset element path, deep-copied, keeping dataset id, format and constructed. Elements appear in path order. Overlapping paths replace rather than duplicate. Same-id datasets stay apart, matched by position.

Behavior changes (for the ChangeLog)

• Data-loss fix. With a dataset field packager, a dataset path whose segment count doesn't match the envelope made setDatasetPath return false. The plain path code then replaced the field: on main, set("104.0xDF02", …) after set("104.0x01.0xDF01", …) with cmfv3.xml turns DE 104 into an ISOMsg, and pack() fails with error packing field 104 (reproduced locally). set/unset now throw (IllegalArgumentException, or ISOException from set(String, ISOComponent)).
• Envelope-less dataset fields numbered above 0x70 got DBM datasets from set(...); they now get TLV.
• TLV elements written by path now record constructed. This is metadata only; TLV packing doesn't use it.
• Unsetting a dataset path removes the element from every dataset with that id, not only the first.
• clone(String...) throws IllegalArgumentException for malformed paths instead of an incidental NumberFormatException/NoSuchElementException.

Testing

• New ISOMsgPathTest (49 tests): the grammar cases, strict walk, dataset descent with and without a packager (cmfv3.xml), mismatched layouts, nested 127.55.x with and without a sub-message packager, XMLPackager round-trip, set() without a packager producing a plain composite, repeated tags (issuer scripts decoded with TLVList) and repeated datasets, the writers including pack/unpack with cmfv3.xml, the mismatch regression, the TLV format fix, and clone with dataset paths.
• ISOMsgTest.testCloneFPathBrokenHierarchy (from ISOMsg: return null from getComponent(String) on a broken hierarchy #791) no longer asserts that clone("55.0x9F26") skips DE 55, since that path is now cloned. The leaf case is still covered.
• Full :jpos test suite: 4,766 tests, 0 failures, 0 errors, 69 skipped (local run, macOS). :jpos:javadoc reports no warnings.
• Other modules were not run.