Resolve tf.TensorShape literals in shape arguments

[khatchad]

Closes wala#789.

tf.TensorShape was unsummarized, so a declared shape wrapped in the constructor form (output_shapes=tf.TensorShape([2])) was present but unparseable and honestly degraded the shape axis to ⊤ while the dtype survived. The constructor is now modeled like TensorSpec: it stores its dims argument on the result, and the shape-argument parser recurses into the stored field at both the top-level and nested positions. The declared-shapes guard flips to full precision: the TensorShape([2]) form composes (2,) through the from_generator transformation chain exactly like the positional and keyword tuple forms.

🤖 Generated with Claude Code

https://claude.ai/code/session_01GoqNGum9YcBd81MuypTWwX

[codecov]

Codecov Report

❌ Patch coverage is 84.61538% with 2 lines in your changes missing coverage. Please review.
✅ Project coverage is 75.52%. Comparing base (51171f2) to head (88ee34f).

Files with missing lines	Patch %	Lines
...bm/wala/cast/python/ml/client/TensorGenerator.java	71.42%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff            @@
##             master     #706   +/-   ##
=========================================
  Coverage     75.51%   75.52%           
- Complexity     4831     4833    +2     
=========================================
  Files           340      340           
  Lines         28133    28142    +9     
  Branches       5254     5257    +3     
=========================================
+ Hits          21245    21254    +9     
- Misses         4781     4782    +1     
+ Partials       2107     2106    -1     

☔ 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.