Drop the high-memory label from WeakConvergence3

[ChrisRackauckas-Claude]

Please ignore this PR until it has been reviewed by @ChrisRackauckas.

WeakConvergence3 (srk_weak_final_non_diagonal.jl) is pinned to a high-memory runner label that no runner in the SciML fleet carries, so the job is never scheduled — it sits queued for 24 h, is expired by GitHub, and holds the whole Sublibrary CI run open behind it (#4030).

The label is not needed. This group's studies pass a reducing output_func to EnsembleProblem and an expected_value to test_convergence:

ensemble_prob = EnsembleProblem(prob;
    output_func = (sol, ctx) -> (h2(sol.u[end][1]), false), prob_func = prob_func)

so each trajectory contributes 8 bytes to the retained result rather than a whole solution object — measured at exactly 8.0 B/trajectory. It is a slow group, not a memory-hungry one.

Measured

Run to completion on an 8 vCPU box, GROUP=WeakConvergence3, JULIA_NUM_THREADS=8:

peak RSS 2.56 GiB
wall     1:26:23
         Testing StochasticDiffEqWeak tests passed

2.56 GiB fits the standard self-hosted pool with room to spare, so point it there. The 300-minute timeout is left alone — at 86 minutes the group genuinely is slow, and that is the real reason it is expensive.

Scope

Only this group. The other four high-memory groups in this file are being measured separately and are not touched here — three of them did not finish inside the two-hour budget I first gave them, so I have no completed figure for those yet and will not unpin on a projection. I have already been wrong once about which groups a memory fix rescues (SROCKC2WeakConvergence turned out unaffected by the test_convergence retention fix in #4056), so these are going one at a time on completed runs only.

---

Links

• Blocked-CI issue: Route CI jobs to runner pools that exist and can hold them #4030
• Memory survey: NoncommutativeConvergence test group grows to >51 GiB with no plateau #4036 (comment)
• Companion PR reducing the genuinely-large groups: Summarise the Monte-Carlo test_convergence ensembles by default (breaking) #4056

🤖 Generated with Claude Code

https://claude.ai/code/session_013WW2jGeoWZVZu7AiscUNSz