Migrate the public API from 1-indexed (Julia convention) to 0-indexed

[shinaoka]

Problem

As a port of QuanticsGrids.jl, this crate inherited Julia's 1-indexed convention for grid indices and quantics digits. In a Rust crate this is a foreign convention, and it propagates: tensor4all-rs's tensor4all-quanticstci exposes 1-indexed &[i64] grid indices on its public surface (interpolation callbacks, evaluate, initial_pivots, cachedata keys) purely because this dependency is 1-indexed, while every other tensor4all-rs crate is 0-indexed. The resulting mixed convention is the single worst user trap in that workspace: a Rust user who has just internalized "sites are 0-indexed here" is betrayed by exactly one crate (documented in tensor4all/tensor4all-rs#363, and re-flagged by the 2026-08 AX audit as the top silent-wrong-answer risk for users and coding agents arriving from ITensors.jl or plain Rust).

quanticstci currently pays for the mismatch with five explicit plus/minus-one conversion sites at the TCI boundary (quantics_tci.rs lines 178, 478, 523, 774, 811 at the audited rev).

Proposal

Migrate the public API to 0-indexed grid indices and quantics digits throughout:

• Grid indices: 0..N instead of 1..=N; the first grid point is [0, 0, ...].
• Quantics digits: 0/1 bit values and 0-indexed digit positions where applicable.
• Prefer usize for index-typed parameters where negative values are impossible, making the convention visible in the type.

Compatibility with the Julia ecosystem is a real constraint, but it belongs at the interop boundary, not in the Rust core: a Julia caller crossing FFI (or a test comparing against QuanticsGrids.jl output) converts once at the boundary. Keeping the Rust core 1-indexed inverts that cost onto every pure-Rust user.

Migration strategy

• This is a breaking change; land it as a version bump with a clear CHANGELOG entry stating the old and new conventions.
• Coordinated downstream update: tensor4all-rs pins this crate by rev, so land the quanticstci adoption (removal of the conversion sites and of its 1-indexed documentation) in the same coordinated change set. Follow-up issue on tensor4all-rs side.
• Provide a conversion note for users porting QuanticsGrids.jl scripts: subtract 1 from grid indices at the call boundary.

Acceptance criteria

• No public API accepts or returns 1-indexed grid indices or digits.
• README and docs state the 0-indexed convention and the QuanticsGrids.jl porting note.
• Doctests and tests assert 0-indexed semantics (first grid point [0, ...]).