[WIP] shell: select task command from scheduler-matched slot label

[zekemorton]

Summary

This PR adds the flux-core shell support needed to select a task command based on the slot alternative
matched by the scheduler.

It complements flux-sched#1550, which records the
matched flexible-slot label in the scheduler’s R metadata. Together, the changes allow a jobspec with
multiple tasks[] entries—each associated with a tasks[].slot label—to run the command corresponding to
the resource alternative selected at scheduling time.

Implementation

• Preserve the complete jobspec tasks array during shell parsing, instead of retaining only tasks[0].
• Continue to use tasks[0] as the default command for ordinary single-task jobspecs.
• Read the current broker rank’s matched slot label from:
  scheduling.graph.nodes[].metadata.ephemeral.slot_label
  in R.
• Resolve job->command to the task whose slot matches that label before tasks are created. This means
  existing consumers of job->command, including task argv construction and the default job name,
  automatically use the selected command.
• Fail clearly if R identifies a slot label for which the jobspec has no task.
• Reject conflicting labels for resources allocated to the same rank, since the shell currently selects one
  command per rank.

Compatibility

Jobs without labeled task alternatives are unchanged:

• Single-task jobspecs continue to run tasks[0].
• Jobs scheduled with an R match format that does not include scheduling metadata, such as rv1_nosched,
  also retain the tasks[0] default.

Testing

Adds unit coverage for multi-task jobspec parsing and command resolution, including:

• default selection of tasks[0];
• NULL and empty labels;
• matching labels selecting the corresponding command; and
• unknown labels returning an error.

Flux-sched companion PR here

[codecov]

Codecov Report

❌ Patch coverage is 56.00000% with 22 lines in your changes missing coverage. Please review.
✅ Project coverage is 83.82%. Comparing base (f4d31d9) to head (7e6350a).
⚠️ Report is 1 commits behind head on master.

Files with missing lines	Patch %	Lines
src/shell/info.c	35.48%	20 Missing ⚠️
src/shell/jobspec.c	89.47%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #7799      +/-   ##
==========================================
- Coverage   83.84%   83.82%   -0.02%     
==========================================
  Files         599      599              
  Lines      102687   102737      +50     
==========================================
+ Hits        86099    86123      +24     
- Misses      16588    16614      +26     

Files with missing lines	Coverage Δ
src/shell/jobspec.c	96.05% <89.47%> (-0.94%)	⬇️
src/shell/info.c	72.53% <35.48%> (-7.10%)	⬇️

... and 11 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[grondo]

We should get an issue open against RFC 20 (Rv1) to add the slot label. This would remove the need for the job shell to reach into an opaque, scheduler-specific scheduling key for this data.

Once the RFC 20 changes are merged, this PR could be updated and perhaps one of the Python schedulers or a test scheduler could be developed to emit the slot label in R for more complete end-to-end testing.