Fix inplace/layout bugs in Numba lapack routines

[ricardoV94]

1. Numba's _copy_to_fortran_order does not seem to do the right thing with vectors and doesn't always copy them.
2. Most methods were forgetting inputs still need to be F-contiguous even if they don't need to be copied.

PR did some tricks to avoid copying C-contiguous arrays when not needed.

Proof of the _copy_to_fortran_order thing:

import numpy as np
import numba
from numba.np.linalg import _copy_to_fortran_order

@numba.njit
def foo(x):
    out = _copy_to_fortran_order(x)
    return out
    
# Actually copies if x is a matrix
x = np.zeros((10, 1))
y = foo(x)
y[0] = np.pi
assert (x == 0).all()

# Doesn't copy if it's a vector
x = np.zeros((10))
y = foo(x)
y[0] = np.pi
assert (x == 0).all()  # AssertionError

[codecov]

Codecov Report

Attention: Patch coverage is 11.76471% with 60 lines in your changes missing coverage. Please review.

Project coverage is 81.98%. Comparing base (a149f6c) to head (689edc2).
Report is 4 commits behind head on main.

Files with missing lines	Patch %	Lines
pytensor/link/numba/dispatch/slinalg.py	11.76%	55 Missing and 5 partials ⚠️

❌ Your patch status has failed because the patch coverage (11.76%) is below the target coverage (100.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1304      +/-   ##
==========================================
- Coverage   82.00%   81.98%   -0.02%     
==========================================
  Files         188      188              
  Lines       48478    48496      +18     
  Branches     8665     8671       +6     
==========================================
+ Hits        39755    39761       +6     
- Misses       6575     6583       +8     
- Partials     2148     2152       +4     

Files with missing lines	Coverage Δ
pytensor/link/numba/dispatch/slinalg.py	44.09% <11.76%> (-0.36%)	⬇️

🚀 New features to boost your workflow:

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

[jessegrabowski]

Small comments that don't matter, but also double-check the logic in posv to make sure we can't do better, since the input is symmetric

[jessegrabowski]

Suggested change

	not (overwrite_b and b_val_c_contig.flags.f_contiguous)
	not (overwrite_b and b_val_c_contig.ndims == 1)

Is more clear? Checking that the c and f flags are both true is not obvious

[ricardoV94]

the col row matrices also have this property

[jessegrabowski]

then .squeee().ndims. I get that checking the other flag is elegant, but it requires detailed knowledge of how all these flags work, which isn't going to obvious to someone reading through these tests for the first time in the future

[ricardoV94]

No, the point is even if you defined it as C-contiguous array, it may still be F-contiguous in which case it will override just like the previous block did. I think this is the most readable test

[jessegrabowski]

I wonder if these bug fixes also fix the error reported here. Not for this PR, just thinking out loud.

[ricardoV94]

I don't know what code you were running to see the error

[jessegrabowski]

Satisfied on the posv point, the rest of the comments can be addressed or not as you see fit