Add Brent's method optimization algorithm

[nyo16]

Summary

This PR adds Scholar.Optimize.Brent, implementing Brent's method for scalar function minimization. This is the second optimization algorithm following the Golden Section merge in #327, continuing the incremental approach discussed in #323.

Brent's method combines the reliability of golden section search with faster convergence from parabolic interpolation, making it the recommended choice for scalar optimization (as noted in https://docs.scipy.org/doc/scipy/tutorial/optimize.html).

Changes

• lib/scholar/optimize/brent.ex - Pure defn implementation of Brent's method
• test/scholar/optimize/brent_test.exs - Comprehensive test suite (11 tests + 2 doctests)
• notebooks/optimize.livemd - Updated to present Brent as recommended method with performance comparisons
• agents.md - Documents best practices from @josevalim's review feedback on Add Golden Section optimization algorithm #327

Features

• JIT/GPU compatible - Implemented as pure defn with while loops and Nx.select
• Same API as Golden Section - Brent.minimize(a, b, fun, opts \ [])
• Significantly faster convergence - ~3-5x fewer function evaluations than Golden Section

Performance Comparison

Function	Bracket	Brent	Golden Section
(x-3)²	[0, 5]	~8 evals	~45 evals
(x-50)²	[0, 100]	~10 evals	~50 evals
sin(x)	[3, 5]	~11 evals	~42 evals

Example

alias Scholar.Optimize.Brent

fun = fn x -> Nx.pow(Nx.subtract(x, 3), 2) end
result = Brent.minimize(0.0, 5.0, fun)

Nx.to_number(result.x) # => 3.0
Nx.to_number(result.fun_evals) # => ~8 (vs ~45 for Golden Section)

Implementation Notes

Following the patterns established in #327 and @josevalim's feedback:

• defn entry point (not deftransform)
• Bounds as explicit function arguments
• while loop with state map (not recursion)
• Nx.select for all conditionals (no Nx.to_number in defn)
• u32 for non-negative counters

I also added agent.md with the learnings from the last 2 PRs. It really helping the models to follow better the best practices.

References

• Brent, R.P. (1973). "Algorithms for Minimization without Derivatives"
• https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.brent.html

[josevalim]

@nyo16 can you please send a separate PR for agents.md? Thank you!

[josevalim]

💚 💙 💜 💛 ❤️