fix: keep circuit and name paired when parsing the find output

[joltcoke]

Problem

FindParams and ebusd_find split the telnet answer on line breaks and commas:

const datas = str.split(/[\r?\n,]+/);
const names = datas[i].split(",");

The first split already flattens ehp,Name\nehp,Other into single tokens, so the
following split(",") finds nothing left and names[1] is always undefined.

• FindParams returns every entry with name undefined and the message names
  sitting in the circuit field, plus one stray entry holding the bare circuit.
• ebusd_find passes undefined as both name and comment to UpdateDP.

The admin compensates by reading item.circuit, which is why the polled datapoint
list still looks right today — that is what kept this hidden.

The character class also contains ? as a literal, so a message name containing
one gets torn apart as well.

Change

Split on line breaks only. That restores the pairing, and the existing empty-entry
filter starts working too — it compared undefined to "", which is false in JS.

In ebusd_find the comment is now taken as the last field instead of the third
one: ebusd quotes it when it contains a comma, and 27 of 354 lines in my
configuration do.

QueriedDPSettings.tsx switches to item.name, which is only correct once the
backend fills it.

Verified against a running ebusd

	before	after
find -c ehp -F circuit,name, 184 lines	148 entries, name populated 0×	147 entries, name populated 147×
stray entry with the bare circuit	yes	gone
ebusd_find → UpdateDP	name and comment undefined	347 calls, all three arguments populated
comments containing a comma	truncated	27 complete, quotes stripped

Note on the build output

This PR touches sources only. Rebuilding here rewrites admin/index_m.html and
src-admin/build/index.html almost entirely because the committed files use CRLF
while the build writes LF — 148 lines of churn for one changed asset reference.
Happy to add the build output if you prefer it in the same PR.