ci: trigger the SharedCore macOS lane on currency-math changes

[bmc08gt]

shared-core-tests.yml runs :libs:currency-math:discrete-curve:iosSimulatorArm64Test, but no pattern in its paths: filter matched libs/currency-math/. A PR touching only the discrete bonding curve never started the workflow, so the curve's Kotlin/Native half and its curve.json / curve_fractional.json parity fixtures went unrun for exactly the changes that could break them. The JVM half was never affected: flipcashTestDebug in the Ubuntu CI lane covers :libs:currency-math:discrete-curve and that workflow has no path filter at all.

The fix is one line: - 'libs/currency-math/**'.

Audit of the rest of the filter

I replayed the filter (GitHub glob semantics, ** crossing separators) over git ls-files for every module named in a Gradle task in the workflow:

Module in a task	Directory	Covered by
:kmp:shared-core	kmp/shared-core	kmp/shared-core/**
:libs:codes:kikcode	libs/codes/kikcode	libs/codes/kikcode/**
:libs:currency-math:discrete-curve	libs/currency-math/discrete-curve	libs/currency-math/** (new)
:libs:encryption:{base58,ed25519,hmac,mnemonic,sha256,sha512}	libs/encryption/*	libs/encryption/**

No file under any of those nine directories is now unmatched. The assembleSharedCoreReleaseXCFramework step adds no further reach: :kmp:shared-core exports exactly those eight modules, and their only cross-module dependencies (ed25519-native from ed25519; utils, hmac, sha256, sha512, base58 from mnemonic) all sit under libs/encryption/.

In the other direction, three paths match the filter without belonging to any task here. All three over-trigger rather than under-trigger, so I've left them:

• libs/encryption/keys/** — a plain Android module with no KMP target. Nothing reachable from this workflow depends on it; its consumers are libs/crypto/solana, libs/models and app modules. This is the one genuinely spurious match.
• libs/currency-math/** outside discrete-curve/ — the parent Android module, added by this change. It depends on discrete-curve rather than the reverse, so changes to it can't alter the curve's test result. Narrowing to libs/currency-math/discrete-curve/** would be tighter; I kept the directory-level pattern to match how libs/encryption/** is written.
• libs/encryption/{ed25519-native,utils}/** — not named in a task, but real dependencies of modules that are, so they belong.

Not changed, but worth recording: the filter doesn't cover build-logic/**, settings.gradle.kts, gradle.properties or gradle/wrapper/**, each of which can change what these modules compile to. That's a wider question about how much the macOS lane should cost, not the same defect.

Other workflows

shared-core-tests.yml is the only workflow in the repo with a paths: filter (grep -rn 'paths:' .github/workflows/), so no other workflow can have this mismatch:

• ci.yml — the Ubuntu lane. on: pull_request with no filter, so it runs on every PR. Its module list also isn't hand-written: flipcashTestDebug in the root build.gradle.kts is built from the project lists settings.gradle.kts derives from unitTestPaths, so a new module joins it by being included in the build.
• publish-shared-core.yml — runs :kmp:shared-core:kmmBridgePublish, workflow_dispatch only.
• maestro.yml — workflow_dispatch and schedule.
• bump-patch.yml, prep-dev.yml, build-fcash2-upload-android.yml, labeler.yml — no fixed module list against a path filter.

.github/labeler.yml is the other glob list in the repo. It has no area: entry covering libs/currency-math/**, so curve PRs go unlabelled, but that's a labelling gap, not a test one — happy to fix it here if you'd rather it rode along.