fix: clear error when a multi-output submodel call has the wrong number of LHS outputs

[bvdmitri]

Closes #310.

Problem

A multi-output submodel call must provide exactly as many outputs on the LHS as there are
interfaces left unspecified on the RHS. When it doesn't, the user got:

MethodError: no method matching make_node!(::GraphPPL.Composite, ::GraphPPL.Model{MetaGraphsNext.MetaGraph{Int64, Graphs.SimpleGraphs.SimpleGraph{Int64}, ...

which says nothing about the actual mistake.

Root cause

Not where #310 suggested. The issue proposed adding the check to prepare_interfaces_multi, but
that function is never reached: the dispatch sites at src/graph_engine.jl:2055/:2060 pass
static(length(rhs_interfaces) + length(lhs_interface)), and the generated make_node! methods are
defined only for the model's own StaticInt{num_interfaces}. A count mismatch means no method
matches at all, so it fails one layer earlier than the issue assumed.

Change

Added a less specific make_node! fallback for Composite nodes with Tuple or NamedTuple LHS and
a generic StaticInt{N}, which reports the mismatch directly:

Node 'two_iface_sub' cannot be called with 2 output(s) on the left-hand side and 2 interface(s)
on the right-hand side, 4 in total.

The number of outputs on the left-hand side must be equal to the number of interfaces of
'two_iface_sub' that are left unspecified on the right-hand side. Currently specified interfaces
are: (:x, :b), but check the documentation to see the specification options.

It is strictly less specific than every generated method (generic in both fform and N), so it only
runs when dispatch would otherwise fail — no ambiguity, no shadowing. Message style follows the
existing single-output validation at :1858-1873 (lazy"...", Node '$(fform)', local n = "\n",
closing pointer to the documentation).

Tests

test/multi_and_zero_output_tests.jl currently pins the bad behaviour with
@test_throws "no method matching make_node!". That assertion is replaced with one on the readable
message, plus two new cases: too few LHS outputs, and the same mismatch in the named-output
(NamedTuple LHS) form.

Verified the existing clear errors for the named form are preserved — a wrong interface name still
matches on arity and reaches prepare_interfaces, so "Interface ':c' does not exist in ..." and the
LHS/RHS conflict error still fire (covered by the existing test items 8 and 9).

Full suite on Julia 1.12: 258/258 test items, all passing.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.73%. Comparing base (4cc7c6c) to head (857ddc1).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #324      +/-   ##
==========================================
+ Coverage   90.72%   90.73%   +0.01%     
==========================================
  Files          16       16              
  Lines        2263     2266       +3     
==========================================
+ Hits         2053     2056       +3     
  Misses        210      210              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.