Add tests for shrinking behavior of generators

[whatyouhide]

Closes #160.

What changed

Removes the long-standing # TODO: test shrinking from StreamDataTest and adds shrinking tests for essentially every shrinkable generator, each placed alongside the generator's existing tests (inside its describe block where one exists, as a sibling test otherwise — same pattern already used by one_of/1 and shuffle/1).

The existing shrink/1 helper (originally taken from the issue) is extended to shrink/2 with an optional predicate: it returns the smallest value the generator shrinks to for which the predicate holds, by running check_all/3 with a property that fails exactly on those values. This enables classic assertions like:

assert shrink(integer(), &(&1 >= 42)) == 42
assert shrink(list_of(integer()), &(length(&1) >= 3)) == [0, 0, 0]

It also includes the higher-order property suggested in the issue: integer(bound1..bound2) shrinks to the smallest absolute value in the range (0 if the range straddles zero, otherwise the bound closest to zero — per the documented behavior, not min(bound1, bound2) as the issue sketch guessed).

Notes for review

• Each test asserts what the generator's own "Shrinking" doc section promises: integer/0 → 0, boolean/0 → false, list_of/2 → [] (respecting :min_length), member_of/1/frequency/1/one_of/1 → earliest entries, optional_map/1 → %{}, filter/2 only shrinks within the predicate, unshrinkable/1/repeatedly/1 visit zero shrink nodes, and so on.
• For float/1, only bound-respecting is asserted: the docs explicitly say floats shrink towards "simpler" values, not smaller ones, and the exact shrink targets are implementation artifacts.
• All shrinks run from :os.timestamp() seeds, so every exact-value assertion was verified deterministic across 50 random seeds before being encoded, and the suite was run repeatedly with different ExUnit seeds to check for flakiness.
• The helper bumps max_runs/max_shrinking_steps to 1000 so predicate-based shrinks reliably find a failing value to shrink from.

[coveralls]

Coverage Report for CI Build 302

Coverage increased (+1.1%) to 94.457%

Details

• Coverage increased (+1.1%) from the base build.
• Patch coverage: No coverable lines changed in this PR.
• No coverage regressions found.

Uncovered Changes

No uncovered changes found.

Coverage Regressions

No coverage regressions found.

---

Coverage Stats

Relevant Lines:	451
Covered Lines:	426
Line Coverage:	94.46%
Coverage Strength:	11758.51 hits per line

---

💛 - Coveralls