[yugabyte#28803] docdb: Canonicalize thread names for perf endpoint

Summary:
The beta `/perf` endpoint generates flamegraphs. The default stack-collapse.pl script combines threads by name, but some of our threads have index numbers. This diff canonicalizes the thread names by truncating at the first numerical character and replacing all numbers with 0, so `prepare_347_wor` becomes `prepare_0`. (The code in this diff was created by @kfranz, I am just putting up the diff.)
Jira: DB-18514

Test Plan:
Checked perf endpoint and the stacks are correctly aggregated across workers now:
{F404101}

The stacks used to look like this, since each worker thread was a separate tower:
{F404103}

Reviewers: hsunder, kfranz

Reviewed By: kfranz

Subscribers: kfranz, ybase

Differential Revision: https://phorge.dev.yugabyte.com/D47496

Author: SrivastavaAnubhav

src/yb/tools/stackcollapse-perf.pl

@@ -284,6 +284,13 @@ sub inline {
 		}
 		($m_pid, $m_tid, $m_period) = ($pid, $tid, $period);
 
+		# Normalize digits in the thread name (comm) to a single 0 and then
+		# truncate at the first digit encountered so names like
+		# "prepare_1_worker" and "prepare_347_wor" both become "prepare_0".
+		if ($comm =~ /^(.*?)(\d)/) {
+			$comm = $1 . "0";
+		}
+
 		if ($include_tid) {
 			$pname = "$comm-$m_pid/$m_tid";
 		} elsif ($include_pid) {