Add memory device option to support shotwise output

[radumarg]

Summary

• Shot results from IonQ for QPU jobs or noisy simulators jobs will be retrieved when user requests it by setting the "memory" argument to True on device. By default memory is False (opt-in behavior).
• will not attempt to retrieve shotwise results for ideal simulation jobs since these are not available on the IonQ servers.
• when the shotwise results will not be present, either because this is an ideal simulation or the user has not requested to be retrieved, and the user requests circuit/tape result other than probabilities, the memory (shotwise) results will be generated in software from retrieved probabilities as it is already done right now.
• when "memory" is False and shots are needed, the current behavior where shots are generated by sampling the probability distribution will take over
• for unlikely 4xx/5xx/RequestException errors on retrieving shots a warning will be generated but job will not fail, shots will be instead generated in software for that particular job

[splch]

Skimmed since this is still draft - just front-loading notes for the next push.

Main structural thing: shotwise_results is set but generate_samples doesn't read it, so qml.sample() output won't actually change yet. And memory=True default looks opposite of the "only when requested" intent. Smaller inline notes below.

[splch]

For the next push: would be good to cover request shape with memory on/off, ideal-sim opt-out, and a round-trip from raw shots into qml.sample() output.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 90.01%. Comparing base (2e61d26) to head (caaf22f).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #178      +/-   ##
==========================================
- Coverage   98.80%   90.01%   -8.80%     
==========================================
  Files           6        6              
  Lines         587      671      +84     
==========================================
+ Hits          580      604      +24     
- Misses          7       67      +60     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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

[splch]

Cross-SDK we settled on memory and off by default. Needed before merge:

1. Rename shotwise → memory everywhere (device kwargs, api_client.py, docstrings, test file).
2. Flip default True → False; update CHANGELOG to say opt-in.
3. Drop the fake-query-param dance — pass a real fetch_shots: bool into refresh_data instead of api_params.pop("shotwise", True).
4. Wrap the shots GET in _retrieve_child_job_shots with try/except + warn, matching qiskit-ionq, so a 5xx doesn't fail .result().

Multi-circuit logic stays — qiskit-ionq is copying it.

[splch]

qiskit-ionq#234 just merged with the same shape — opt-in, per-child fetch, ideal-sim skip. One behavior issue worth blocking on, one shape concern, a few nits inline.