bump: version 0.11.0 → 0.12.0

[spMohanty]

Bumps all 8 version locations 0.11.0 → 0.12.0 and writes the release notes.

The minor is forced. cz bump reports PATCH from these commits — every one since v0.11.0 is fix/test/docs/style/perf, with no feat: and no BREAKING CHANGE. The release nonetheless raises what the grader charges for several operations and opens a new phase, which a patch digit would understate.

The notes are assembled, not generated. update_changelog_on_bump = true, so cz bump rewrites ## Unreleased into a flat bullet list — which would have dropped the measured billing-impact prose entirely. The prose for this release lived in three disjoint places and all three are now merged into one section:

• the hand-written section on the abandoned release-v0.12.0 branch (fix(billing): apply NEP 50 weak promotion to where and insert operands #238–fix(billing): close the out= dtype bypass, NaN q leak, and nan* pass mis-modelling #243), which was cut at fix(billing): close the out= dtype bypass, NaN q leak, and nan* pass mis-modelling #243 and never opened as a PR;
• main's ## Unreleased (fix(billing): charge the float loop six bool-returning ops compute in #247–perf(einsum): memoize the symmetry-aware path rebuild instead of redoing it per call #251);
• perf(client): stop charging every payload leaf twice on dispatch #244 and fix(billing): price ldexp on its mantissa loop, and stop charging for calls NumPy refuses #245, whose prose existed nowhere.

Billing-impact items are split into costs more and costs less rather than one undifferentiated list, with a separate no billed amount changes group for the neutral work. The header states plainly that the release ships as a new version opening a new phase, so no submission is re-evaluated against it. cz's generated commit lists are kept below the prose, per the existing format.

Both stale release branches are deleted (release-v0.11.1 f55c72611d, release-v0.12.0 5a3f4e56ed, recorded here in case anything needs recovering).

Verified: all 8 version locations at 0.12.0, three lockfiles refreshed and consistent, 10,286 tests pass, ruff and pyright clean.

No tag is pushed by this PR. Tagging triggers the PyPI publish workflow and is a separate, deliberate step.