Fix max_codes=0 (unlimited) in the binary IVF count scanner

[rioyu123]

Related to #5555, reported by @leemeii.

Summary

With IndexBinaryIVF.use_heap=false, max_codes=0 currently stops the count
scanner before it examines any codes. Searches using the default unlimited
budget therefore return only missing-result placeholders.

This patch gives the count scanner the same zero-budget normalization as the
heap scanner. Index-level defaults and per-search zero overrides now
mean unlimited scanning; positive budgets retain their existing behavior.

Coordination

I asked in #5555 whether the original reporter already has a patch in progress.
I'm sharing this small implementation for review, without
assuming that coordination is settled. If there is work underway, I'm happy
to coordinate and avoid overlapping submissions.

Validation

• Added a deterministic regression with both inverted lists populated. It
  compares the count and heap scanners with IndexBinaryFlat for the default
  zero budget and an explicit zero override, and checks that a positive budget
  still limits the results.
• The new test fails on the unchanged base and passes with this patch.
• Both binary IVF tests in test_lowlevel_ivf.cpp and the existing TPOB.IVF
  parameter-override test pass against the CPU-only source build.
• git diff --check and the changed-lines git clang-format check pass.

The full low-level IVF test file reached a 60-second time limit in the existing
IVFRQ test, after the first eight tests passed. Full-suite, GPU, and Python
binding validation were not completed.