Describe the bug

The pull request #546, specifically the commit 6035ffb, introduced a regression that results in a build failure when the template (i.e. #[derive(Template)]) is actually generated from within a user defined macro, that happens to live in a different location than the file where the macro is expanded into.

To Reproduce

Here is the short version:

• have a macro defined in a file that calls #[derive(Template)];
• have a module that uses the above macro;
• if the macro file doesn't live in the same folder as the module file, then the computed relative path by rel_path for the include_bytes (in the above mentioned commit) is wrong;
• this is because the path is computed relative to the macro file, not relative to the final module file where the #[derive(Template)] is actually expanded into, and which the rustc will examine;

When can such a mismatch happen:

• the macro file and the module file live in different folders, at different depths, but inside the same crate; (as the minimal reproducible example below;)
• when the macro file is part of another crate, and the module file is part of another crate;

cargo init --lib askama-bugs
cd ./askama-bugs
cargo add askama

mkdir ./templates
touch ./templates/empty.txt

cat > ./src/m.rs << 'EOS'
#[ macro_export ]
macro_rules! define_template {
    () => {
        #[ derive (askama::Template) ]
        #[ template (path = "empty.txt") ]
        struct Empty {}
    }
}
EOS

mkdir ./src/a
cat > ./src/a/a.rs << 'EOS'
crate::define_template!();
EOS

cat >| ./src/lib.rs << 'EOS'

#[ path = "./a/a.rs" ]
mod a;

mod m;

EOS

cargo check

error: couldn't read `src/./a/../templates/empty.txt`: No such file or directory (os error 2)
 --> src/m.rs:4:20
  |
4 |         #[ derive (askama::Template) ]
  |                    ^^^^^^^^^^^^^^^^
  |
 ::: src/./a/a.rs:1:1
  |
1 | crate::define_template!();
  | ------------------------- in this macro invocation

Askama version
v0.15.4

Rust version
1.94.0