docs: tutorials for parametric args, QMC, VEGAS error and NN training

[gomezzz]

What

Tutorial documentation for the 0.6 features and two long-open doc requests. Adds four sections to docs/source/tutorial.rst:

• Passing parameters to the integrand — the args= argument as a lambda-free way to integrate parametric integrands (Integrate function with parameters #187).
• Quasi-Monte Carlo with Sobol points — rng=Sobol(...) with MonteCarlo.
• VEGAS error / minimum-N — return_error=True → VEGASResult, plus guidance on choosing N.
• Training a neural network through an integral — a runnable example minimizing an integrated loss with gradients flowing through the quadrature (Add more complex examples to tutorials showcasing differentiability #125).

Notes

• Merge order: documents features added in feat: add Sobol quasi-Monte Carlo sampler #244 (Sobol), feat: add opt-in VEGAS error reporting (return_error) #245 (VEGAS return_error) and feat: forward integrand args via integrate(..., args=...) #246 (args=), so this should merge after them. The docs build is unaffected either way (RST code blocks aren't executed; the Sobol cross-reference resolves once feat: add Sobol quasi-Monte Carlo sampler #244 lands).
• The NN example was run and converges (integrated L2 error 0.66 → 0.07).

Test plan

• sphinx-build -W -b html docs/source builds clean

Closes #125, #186. Roadmap F4.

[github-actions]

Coverage Report

File	Stmts	Miss	Cover	Missing
torchquad
__init__.py	25	2	92%	63–64
torchquad/integration
base_integrator.py	49	7	86%	28, 71–74, 78, 113, 117, 122
boole.py	30	3	90%	53, 72–75
gaussian.py	51	3	94%	83, 107, 127
grid_integrator.py	97	3	97%	49, 163, 267
integration_grid.py	49	4	92%	64–65, 132, 135
monte_carlo.py	93	3	97%	77, 99, 228
rng.py	60	7	88%	71–74, 93–94, 102
simpson.py	29	3	90%	51, 70–73
utils.py	131	11	92%	35–36, 102, 173, 177, 184, 196, 198, 204, 299, 306
vegas.py	161	4	98%	100–101, 187, 334
vegas_map.py	103	8	92%	247–257
torchquad/utils
deployment_test.py	192	63	67%	53–55, 65–66, 69–70, 78–79, 91–93, 164, 175–176, 180–181, 183–184, 196–199, 211–213, 221, 232–234, 237, 247, 250, 253, 256–259, 267–275, 280–288, 297–306, 316–323
enable_cuda.py	15	6	60%	17–22
set_log_level.py	14	2	86%	45–46
set_precision.py	34	9	74%	35–38, 50–51, 70–72
TOTAL	1229	138	89%

Tests	Skipped	Failures	Errors	Time
92	0 💤	0 ❌	0 🔥	2m 24s ⏱️

[gomezzz]

Claude: Thanks — great catches, both RST defects fixed in 5a0cfd0:

1. Label anchor (Important): moved .. _multi_dim_integrand: back down to immediately precede the Multidimensional/Vectorized Integrands heading, so the :ref: from the vectorized-quadrature section resolves to the right anchor again.

2. Inline math (Important): switched \theta/\int/\,/\sqrt{N} to single backslashes so MathJax renders them (the double-backslash cases elsewhere are inside a .. code:: Python docstring, so those stay).

3. Undefined integrand: the Sobol example now defines a small smooth_integrand so it's self-contained like the others.

4. Merge order: acknowledged — this documents Sobol (feat: add Sobol quasi-Monte Carlo sampler #244), return_error (feat: add opt-in VEGAS error reporting (return_error) #245) and args= (feat: forward integrand args via integrate(..., args=...) #246), so it should merge after them. Flagged in the PR description.

sphinx-build -W still builds clean.