Move the MIKE+ station lookup out of #702

[jpalm3r]

#702 carries a MIKE+ database lookup that places measured timeseries in the network:
_MikePlusStationResolver (~290 lines in obs.py), _observations_from_mikeplus, the db=
and source= arguments on NodeObservation.from_multiple and
ReachObservation.from_multiple, tests/test_mikeplus.py (558 lines), and a section of the
user guide. None of it is needed for the mikeio1d move #702 is about (ADR-013, Phase 2), and
it is not ready to ship on its own terms.

What needs more thought:

• No fixture. Every test builds its own sqlite database with build_db, so the schema
  assumptions are checked only against our own construction of them. Nothing here has been
  read against a database MIKE+ wrote.
• The locationtype codes are magic numbers. _NODE_TYPES = {8, 12},
  _LINK_TYPES = {9}. Any other code is dropped and named only on the failure path. No
  source says these are the codes, or that they are stable across MIKE+ versions.
• resitemname is split on ; and the first field taken. An undocumented encoding.
• tsfilename holds a Windows path and is matched on basename alone, case-folded. Two
  result files with the same name in different folders collide; source= is the only escape.
• The nine columns across m_Station and m_Measurement are validated against one
  layout. The error says "The database layout is not the one modelskill expects" without
  recording which layout that is.
• assetname becomes the observation name only when it is unique across the selection.
  The names a user gets depend on what else is in the database.

from_multiple(db=...) is public API, so whatever shape lands ships with 1.4.0.

Extract it onto its own branch, off #702 or off main once #702 lands.