perf: vectorize a loop-based matrix constructor

[vprusso]

A few small matrix constructors build their result with a Python for loop over a preallocated np.zeros/np.empty, where numpy vectorization would be faster and clearer.

What to do

Pick ONE of the following and replace its loop with a vectorized construction:

• toqito/matrices/standard_basis.py
• toqito/matrices/gen_pauli_z.py
• toqito/matrices/entangled_subspace.py
• toqito/perms/perfect_matchings.py

Requirements

• The output must be bit-for-bit identical to the current implementation. Keep the existing test, and add a case comparing the new output to the previous one if helpful.
• Include a measured speedup on a representative size (a short timeit snippet in the PR description is enough).
• Do not change the function's signature or behavior.

This one needs some numpy comfort, so it is a good pick if you want a slightly meatier first PR.

Verifying

uv run pytest toqito/matrices/tests toqito/perms/tests

[vprusso]

Two of the candidate functions are now done: gen_pauli_z (#1893) and standard_basis (#1890). Reopening because the other two candidates are still available for anyone who wants a first PR:

• toqito/matrices/entangled_subspace.py
• toqito/perms/perfect_matchings.py

Same requirements as before: bit-identical output (keep/extend the test) and a measured speedup.

[vprusso]

Closing this in favor of per-function issues:

• entangled_subspace: perf: vectorize the entangled_subspace construction #1894
• ghz and w_state: perf: vectorize the ghz and w_state constructors #1895

Note on the last candidate: perfect_matchings is recursive combinatorial generation (it calls itself and stacks results, with output size growing as (2n-1)!!), so it is not a clean vectorization target and has been dropped from this line of work.

Already done: gen_pauli_z (#1893) and standard_basis (#1890).