run: accept --running_platform so platforms get separate series

[paulorsousa]

Problem

running_platform is already:

• part of the time-series key, in get_ts_metric_name (utils/utils.py)
• a label, in get_project_ts_tags (utils/remote.py)
• threaded through common_exporter_logic and timeseries_test_sucess_flow
• filterable in compare, via --running_platform (compare/args.py)

Only the run entry points never supply it. run_remote.py and run_local.py both
call timeseries_test_sucess_flow without it, and no run argument exists to set
it, so in practice it is always None.

The consequence is that benchmark runs on different hardware share a single
series. Tagging cannot substitute for this: check_rts_labels applies TS.ALTER
whenever incoming labels differ from the key's, so labels describe the series
and the most recent push wins. Provenance cannot be recovered per sample.

Observed while building regression coverage for RediSearch: an m7i.8xlarge VM
baseline and an m7i.metal-24xl run against the same commit read back as one
8-sample series. Its confidence interval converged happily, around a mean that
described neither platform. Nothing on the surface looked wrong.

Change

Adds --running_platform to common_run_args, beside --architecture — its
closest analogue, since both describe the machine the benchmark ran on and both
belong in the key. Both entry points forward it. 13 lines, no new plumbing.

Compatibility

Leaving the flag unset produces byte-identical keys, so no existing series move
and no historical data is invalidated. Setting it starts a new series per
platform, which is the intent.

unset : ci.benchmarks.redislabs/by.hash/circleci/O/R/t/oss-standalone/<sha>/Ops/sec
m7    : ci.benchmarks.redislabs/by.hash/circleci/O/R/t/redisearch-m7/oss-standalone/<sha>/Ops/sec
metal : ci.benchmarks.redislabs/by.hash/circleci/O/R/t/redisearch-m7i-metal/oss-standalone/<sha>/Ops/sec

Verification

Checked directly against the key-building helpers:

• m7 and m7i-metal produce different keys
• unset produces a key byte-identical to before the change
• the label is present when set and absent when unset
• --running_platform parses in both run-remote and run-local

Caveat for reviewers: this was verified by exercising the pure key-building
helpers with third-party imports stubbed, not by running the project's own test
suite, which was not available in the environment used.

Deliberately out of scope

export_redis_metrics is untouched. It builds its own key prefix for server-side
metrics, which are not used for regression gating, and changing that prefix would
relocate a separate metric family.

Note for consumers

Callers pinning redisbench_admin>=0.12.29 from PyPI will not have this flag
until it is released; passing it against a released version fails with
unrecognized arguments.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 47.67%. Comparing base (aca22be) to head (449e375).

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #548      +/-   ##
==========================================
+ Coverage   47.63%   47.67%   +0.03%     
==========================================
  Files          74       74              
  Lines        8917     8918       +1     
==========================================
+ Hits         4248     4252       +4     
+ Misses       4669     4666       -3     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

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