Remove dtype promotion in pauli rep of scalar products with tensorflow

[Qottmann]

fix for https://github.com/PennyLaneAI/pennylane/issues/5245

[sc-57300]

[github-actions[bot]]

Hello. You may have forgotten to update the changelog! Please edit doc/releases/changelog-dev.md with:

• A one-to-two sentence description of the change. You may include a small working example for new features.
• A link back to this PR.
• Your name (or GitHub username) in the contributors section.

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 99.67%. Comparing base (d2568ad) to head (49d9553). Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5246      +/-   ##
==========================================
- Coverage   99.68%   99.67%   -0.01%     
==========================================
  Files         399      399              
  Lines       36871    36583     -288     
==========================================
- Hits        36754    36465     -289     
- Misses        117      118       +1     

:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.

[timmysilv]

@Jaybsoni do you remember why this was added in the first place? You first made a change here, then I had some issue so I just explicitly set it to complex128. if there's no hidden issue with this removal, then lgtm! just wanna make sure with the experts here.

To gather some context:

• this is the conversation regarding why I changed it to be explicit
• #4249 fixed a related issue in the past. TF is delicate, but I assume we have the appropriate tests already in place, so the green CI is easing my mind a bit

[albi3ro]

I also thought this was done deliberately to fix errors rising due to tensorflow conversion issues. But tests seem to be passing so I guess we're ok?

[Qottmann]

I think hitting a problem with dtypes such as

import tensorflow as tf
qml.operation.enable_new_opmath()
a = tf.constant(0.5)
b = tf.constant(0.5j)

x = qml.s_prod(a, X(0))
y = qml.s_prod(b, Y(0))

res = x @ y

TypeError: Cannot convert 1j to EagerTensor of dtype float

when using tensorflow is a user error because with tf one needs to set the correct dtypes to start with:

import tensorflow as tf
qml.operation.enable_new_opmath()
a = tf.constant(0.5, dtype=tf.complex128)
b = tf.constant(0.5j)

x = qml.s_prod(a, X(0))
y = qml.s_prod(b, Y(0))

res = x @ y

(this was the only test I had to change to make this fix work)

In general, we shouldnt force the dtype promotion wherever possible I'd say

[Qottmann]

Can I merge to master right now or should I wait for decoupling the release candidate branch?