Suggest to replace `int_as_float.log{|2|10}().floor() as u32` with `int.ilog{|2|10}()`

What it does

Search for redundant $\lfloor \log_b n \rfloor$ for a integer $n$.

• $\lfloor \log_5 25 \rfloor$ can be written as 25_i32.ilog(5) instead of (25_i32 as f64).log(5.0).floor() as u32.
• $\lfloor \log_2 8 \rfloor$ can be written as 8_i32.ilog2() instead of (8_i32 as f64).log2().floor() as u32.
• $\lfloor \log_{10} 100 \rfloor$ can be written as 100_i32.ilog10() instead of (100_i32 as f64).log10().floor() as u32.

ilog{|2|10} are const fn, but log{|2|10} and floor are not.

const A: u32 = (8_i32 as f64).log2().floor() as u32; // compilation error
const B: u32 = 8_i32.ilog2(); // ok

Advantage

• Simplify.
• Can be used in the const contexts.

Drawbacks

MSRV is 1.67.0.

Example

let x: i32 = 42;
let a = (x as f64).log(42.0).floor() as u32;
let b = (x as f64).log2().floor() as u32;
let c = (x as f64).log10().floor() as u32;

Could be written as:

let x: i32 = 42;
let a = x.ilog(42);
let b = x.ilog2();
let c = x.ilog10();

Start by searching the rust-clippy lint implementations and tests for existing numeric-expression lints and the patterns shown in the issue. Define completion as recognizing the redundant log-and-floor forms, suggesting the corresponding ilog, ilog2, or ilog10 call, and covering the examples and const-context motivation with tests.