Add manual_min_max lint

[Souradip121]

Adds the manual_min_max lint (complexity, warn-by-default).

Detects manual if-else implementations of Ord::min and Ord::max and
suggests using those methods directly.

Closes #16884

What the lint catches

// Before
let _ = if a > b { b } else { a };
let _ = if a < b { a } else { b };
let _ = if a > b { a } else { b };

// After
let _ = a.min(b);
let _ = a.min(b);
let _ = a.max(b);

All eight variants are covered: both strict (>, <) and non-strict
(>=, <=) comparisons, and both orderings of the then/else branches.
Compound expressions are correctly parenthesized:

// Before
let _ = if a + 1 < b { a + 1 } else { b };

// After
let _ = (a + 1).min(b);

Conservative conditions (no false positives)

The lint is skipped when:

• The expression is inside a macro expansion
• The expression is in a const context (Ord::min/Ord::max are not yet const-stable)
• Either operand has side effects (function calls, etc.) — handled implicitly by eq_expr_value which uses SpanlessEq::deny_side_effects
• The type does not implement Ord — this excludes f32/f64, which only implement PartialOrd and whose min/max methods have different NaN semantics

Why not extend manual_clamp?

manual_clamp uses MaybeIncorrect applicability (clamp can panic) and
requires const-evaluatable bounds. These two-way patterns never panic and
have no such constraint, so MachineApplicable is appropriate. A separate
lint keeps the two concerns independent.

changelog: [manual_min_max]: new lint to detect manual implementations of Ord::min and Ord::max using if-else expressions.

[rustbot]

r? @llogiq

rustbot has assigned @llogiq.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

Why was this reviewer chosen?

The reviewer was selected based on:

• Owners of files modified in this PR: 7 candidates
• 7 candidates expanded to 7 candidates
• Random selection from Jarcho, dswij, llogiq, samueltardieu

[github-actions]

Lintcheck changes for 26a71c2

Lint	Added	Removed	Changed
clippy::manual_min_max	1	0	0

This comment will be updated if you push new changes

[ronnodas]

I think this should be b.min(a) since in the equality case the result of if a < b { a } else { b } is b. Of course this doesn't matter for i32s so maybe it makes sense to add tests with types that have fields that are ignored by Ord? Note that in case of ties Ord::min returns the first argument and Ord::max returns the second.

View changes since the review

[Souradip121]

@ronnodas can you check now, just made another commit.

[ronnodas]

I haven't looked at the code at all, just the tests, but I believe the max cases are flipped now, a.max(b) returns b if a.cmp(b) == Ordering::Equal. The PR description above should also be updated (the second example in the "What the lint catches section") for posterity.