For Cachegrind, aggregate all jemalloc functions into a single entry.

Because a single entry like this:
```
120,365,108 (8.7%)  <all-jemalloc-files>:<all-jemalloc-functions>
```
is much more helpful than dozens of entries like this:
```
1,900,760 (0.9%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/include/jemalloc/internal/rtree.h:free
 9,296,168 (0.7%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/include/jemalloc/internal/jemalloc_internal_inlines_c.h:malloc
 7,926,636 (0.6%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/src/jemalloc.c:free
 7,905,743 (0.6%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/include/jemalloc/internal/cache_bin.h:free
 7,834,577 (0.6%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/include/jemalloc/internal/cache_bin.h:malloc
 5,425,997 (0.4%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/src/arena.c:_rjem_je_arena_ralloc
 4,268,616 (0.3%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/src/jemalloc.c:do_rallocx
 3,539,679 (0.3%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/src/arena.c:_rjem_je_arena_ralloc_no_move
 3,093,308 (0.2%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/include/jemalloc/internal/sz.h:malloc
 2,393,337 (0.2%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/src/arena.c:_rjem_je_arena_cache_bin_fill_small
 2,371,694 (0.2%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/include/jemalloc/internal/rtree.h:tcache_bin_flush_edatas_lookup.constprop.0
 2,309,382 (0.2%)  /home/njn/dev/rust2/build/x86_64-unknown-linux-gnu/stage0-rustc/x86_64-unknown-linux-gnu/release/build/jemalloc-sys-1e20251078fe5355/out/build/src/jemalloc.c:malloc
...
```
(And in reality, those jemalloc function entries are interleaved with
many non-jemalloc entries, making them even harder to read.)

Author: nnethercote

collector/src/execute/profiler.rs

@@ -6,7 +6,7 @@ use std::io::BufRead;
 use std::io::Write;
 use std::path::Path;
 use std::process::Command;
-use std::{fs, process};
+use std::{fs, io, process};
 
 // Tools usable with the profiling subcommands, and named on the command line.
 #[derive(Clone, Copy, Debug, PartialEq, clap::ValueEnum)]
@@ -226,7 +226,46 @@ impl<'a> Processor for ProfileProcessor<'a> {
                 let cgout_file = filepath(self.output_dir, &out_file("cgout"));
                 let cgann_file = filepath(self.output_dir, &out_file("cgann"));
 
-                fs::copy(&tmp_cgout_file, &cgout_file)?;
+                // It's useful to filter all `file:function` entries from
+                // jemalloc into a single fake
+                // `<all-jemalloc-files>:<all-jemalloc-functions>` entry. That
+                // way the cost of all allocations is visible in one line,
+                // rather than spread across many small entries.
+                //
+                // The downside is that we don't get any annotations within
+                // jemalloc source files, but this is no real loss, given that
+                // jemalloc is basically a black box whose code we never look
+                // at anyway. DHAT is the best way to profile allocations.
+                let file = fs::File::open(&tmp_cgout_file)?;
+                let reader = io::BufReader::new(file);
+                let mut cgout = fs::File::create(&cgout_file).unwrap();
+                let mut in_jemalloc_file = false;
+
+                // A Cachegrind profile contains `fn=<function-name>` lines,
+                // `fl=<filename>` lines, and everything else. We just need to
+                // modify the `fn=` and `fl=` lines that refer to jemalloc
+                // code.
+                for line in reader.lines() {
+                    let line = line?;
+                    if line.starts_with("fl=") {
+                        // All jemalloc filenames have `/jemalloc/` or
+                        // something like `/jemalloc-sys-1e20251078fe5355` in
+                        // them.
+                        in_jemalloc_file = line.contains("/jemalloc");
+                        if in_jemalloc_file {
+                            writeln!(cgout, "fl=<all-jemalloc-files>")?;
+                            continue;
+                        }
+                    } else if line.starts_with("fn=") {
+                        // Any function within a jemalloc file is a jemalloc
+                        // function.
+                        if in_jemalloc_file {
+                            writeln!(cgout, "fn=<all-jemalloc-functions>")?;
+                            continue;
+                        }
+                    }
+                    writeln!(cgout, "{}", line)?;
+                }
 
                 let mut cg_annotate_cmd = Command::new("cg_annotate");
                 cg_annotate_cmd