Multiple soundness issues in lexical

Summary

`lexical` contains multiple soundness issues:

1. [Bytes::read() allows creating instances of types with invalid bit patterns](https://github.com/Alexhuszagh/rust-lexical/issues/102) 1. [BytesIter::read() advances iterators out of bounds](https://github.com/Alexhuszagh/rust-lexical/issues/101) 1. [The `BytesIter` trait has safety invariants but is public and not marked `unsafe`](https://github.com/Alexhuszagh/rust-lexical/issues/104) 1. [`write_float()` calls `MaybeUninit::assume_init()` on uninitialized data, which is is not allowed by the Rust abstract machine](https://github.com/Alexhuszagh/rust-lexical/issues/95) 1. [`radix()` calls `MaybeUninit::assume_init()` on uninitialized data, which is is not allowed by the Rust abstract machine](https://github.com/Alexhuszagh/rust-lexical/issues/126)

The crate also has some correctness issues.

## Alternatives

For quickly parsing floating-point numbers third-party crates are no longer needed. A fast float parsing algorithm by the author of `lexical` has been [merged](https://github.com/rust-lang/rust/pull/86761) into libcore.

For quickly parsing integers, consider `atoi` and `btoi` crates (100% safe code). `atoi_radix10` provides even faster parsing, but only with `-C target-cpu=native`, and at the cost of some `unsafe`.

For formatting integers in a `#[no_std]` context consider the [`numtoa`](https://crates.io/crates/numtoa) crate.

For working with big numbers consider `num-bigint` and `num-traits`.