New locks in message.c to eliminate contention on content lock

[mgoulish]

Performance changes for flimflam workloads after this change:

               Before        After                      Change

Builtin 3.98 4.04 G Bits/s + 1.5 %

Iperf3 7.87 7.64 G Bits/s - 2.9 %

H2load 8.36 9.00 K Ops/s + 7.1 %

[kgiusti]

This is an excellent patch! See inline for requested changes.
The system_test_two_routers failures are unrelated to this change (seen on main as well).

[mgoulish]

Oh! I had a note about that! And I left it in the other room.

…

On Wed, Mar 5, 2025 at 2:05 PM Ted Ross ***@***.***> wrote: ***@***.**** commented on this pull request. ------------------------------ In src/message_private.h <#1752 (comment)> : > @@ -74,7 +74,12 @@ typedef struct { // typedef struct { - sys_mutex_t lock; + sys_mutex_t lock, + producer_lock, // These locks prevent either side from activating These should have more descriptive, less misleading names. {producer|consumer}_activation_lock — Reply to this email directly, view it on GitHub <#1752 (review)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AFFPY3KU35334YD6D7EU3WT2S5DIXAVCNFSM6AAAAABYMW6YNSVHI2DSMVQWIX3LMV43YUDVNRWFEZLROVSXG5CSMV3GSZLXHMZDMNRSGIZDCNRQGQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[kgiusti]

LGTM - verified that the test failures are due to an unrelated proton issue.
Approval conditional to resolving @ted-ross requests