Clarify dare3/lqr3 truncation semantics and add tests

[baggepinnen]

Summary

• Expand dare3 / lqr3 docstrings to spell out the difference between full=true and the default. The default dare3 returns J*(x, u_prev=0) — the optimal cost-to-go assuming u(k-1)=0 — not the optimal cost-to-go for arbitrary u_prev. The default lqr3 gain is similarly the augmented optimal gain with the u_prev columns dropped, and is therefore suboptimal as a feedback law. Both forms remain useful (e.g. as an MPC terminal cost / feedback when the controller does not track u_prev); the docs now say so explicitly.
• Add three tests in test/test_lqg.jl: Q3 → 0 collapses to standard discrete lqr; the matrix-form entry point dare3(A, B, Q1, Q2, Q3) matches the state-space form; the augmented Riccati residual is satisfied by the full=true solution.

No behavior change — docs and tests only.

Test plan

• julia --project -e 'using Pkg; Pkg.test()' passes
• Inspect the rendered docstrings for dare3 / lqr3

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.77%. Comparing base (c7bdec7) to head (30ab968).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #143      +/-   ##
==========================================
- Coverage   91.81%   91.77%   -0.04%     
==========================================
  Files          20       20              
  Lines        3065     3065              
==========================================
- Hits         2814     2813       -1     
- Misses        251      252       +1     

Flag	Coverage Δ
unittests	91.77% <ø> (-0.04%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 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.