Remove dead code, fix correctness bugs, implement sign_rev

[dpsanders]

Summary

• Delete dead / broken code: src/decorated.jl (references DecoratedInterval which no longer exists in IntervalArithmetic v1), src/pave.jl (pre-Julia-1.0 syntax), src/powers.jl (square!/cube! were unused and the cube_pos implementation is explicitly marked "not rigorous"), examples/examples.jl (references a non-existent square function), and a stray test/Non1788tests/script.jl. Also remove unused mutating helpers exp!, log!, asin!.
• Fix correctness bugs:
  • asin_rev was ignoring its input x argument. Example: asin_rev([0,1], [0,0.5])[2] returned [0, 0.841]; now correctly returns [0, 0.5].
  • max_rev/min_rev now return all-empty when a, b, or c is empty, and guard interval(lo, hi) construction against lo > hi via a small helper. Removes ~9 of the ~10 ill-formed-interval warnings that test runs were emitting.
• Implement sign_rev. The previous body was commented out with broken syntax while still exported. New implementation: sign(x) ∈ {-1, 0, 1}, so the preimage of a is the union of half-lines / {0} depending on which of those values lie in a.
• Docstring cleanup: "byt default" → "by default" (10 sites); cosh_rev/tanh_rev docstrings previously said "Reverse square root"; add default x = entireinterval(b/a) to pow_rev1/pow_rev2, matching what their docstrings already claimed.

Net: +111 / −289 lines across 14 files (5 deleted, 1 added, rest modified).

Test plan

• Pkg.test() passes: 1355 pass (was 1344), 241 broken (same baseline as before)
• 11 new sign_rev tests added in test/Non1788tests/sign.jl covering empty input, empty output, singleton cases (+1, -1, 0), two-value combinations, and restriction by a bounded b.
• Two Non1788 extrema tests that asserted the buggy min_rev/max_rev empty-a behavior were updated to expect emptyinterval(); flagged in the review as incorrect expectations.
• Ill-formed-interval warnings during the test run: reduced from ~10 to 1.

Not addressed here (follow-ups)

• sin!/cos!/tan! are kept because sin_rev/cos_rev/tan_rev route through them via periodise. Making them private / inlining is a larger refactor.
• div_rev does not yet handle 0 ∈ a with extended_div (analogous to what mul_rev already does).
• min_rev/max_rev logic beyond the empty-input guard wasn't rewritten; the existing branch logic is ad-hoc and doesn't achieve tight enclosures in all cases. Recommended to revisit against Goualard (2008, hal-00288457).

🤖 Generated with Claude Code