Collapse jax build/test double-counting by normalizing jax_ref to bare version

[HereThereBeDragons]

Summary

We fix a double-counting bug in the release status.json where every jax build
and test cell is counted twice. The cause is that one jax version reaches the
status pipeline under two spellings: the build matrix names cells by git ref
(rocm-jaxlib-v0.11.0), while the release orchestrator and test dispatch inputs
use the bare version (0.11.0). Left un-normalized, the two spellings key two
distinct variants that never merge, doubling the jax counts.

We normalize every jax ref to the bare version at the two points a ref enters a
variant key, so the two spellings collapse to a single (py, version) cell. On
the affected release this reduces the jax variant count from 30 to 15.

Change

• Add _normalize_ref(axis_key, ref), called uniformly for every fan-out axis.
  Today only jax needs it: a jax_ref value has the rocm-jaxlib-v prefix
  stripped, and the torch axis is a pure passthrough.
• Wire it into both variant-key entry points: _variants_from_jobs
  (job-name parsing) and _variants_from_inputs (dispatch-input parsing).

We strip the prefix rather than add it because stripping is idempotent on
already-bare refs and never mangles a non-version ref such as a branch name.

Tests

• A jax cell derived from a prefixed build job and a bare-version test input now
  collapses to one variant with a single key.
• A torch ref (including slashes) passes through normalization verbatim,
  confirming the torch axis is unaffected.

Scope

The fix applies to future releases only. The already-frozen
20260820/status.json is not retroactively corrected.