Batch per-location reads in from_res1d instead of one to_dataframe() per location

[jpalm3r]

Follow-up from #679.

_load_res1d_network reads each selected location separately: _simplify_colnames calls
node.to_dataframe() per node and per boundary gridpoint, and Res1DReach does the same per
populated gridpoint. Each call crosses the pythonnet/.NET boundary on its own.

mikeio1d exposes a batched form, res.read(queries), taking a list of TimeSeriesId. A test user
measured ~0.5 s for 50 sensors that way, versus a per-trial cost of 18–25 s for a full build today.

Worth checking whether routing the selected locations through one batched read is faster than the
current per-location reads, and how it interacts with the quantities filter added in #679.

[jpalm3r]

#685 fixes a related but distinct waste (a node was read once per reach endpoint even though only the first copy is kept), not the batched res.read(queries) this issue asks for. Still open.