Task: Migrate market_series qualifier/point_id generation onto oresmd

Grepped every real write site of .qualifier =/=point_id = across =ores.marketdata, ores.synthetic, ores.ore, ores.qt:

• FX: ores.synthetic.service/feed_controller.hpp builds ore_key via oresmd_projections::to_quote_key (migrated in the ore_key task), then a local parse_ore_key-style helper (two near-identical copies in this one file, lines ~268 and ~485) splits that already-oresmd string on '/' into series_type=/=metric=/=qualifier for market_series lookups/creation. ores.marketdata.service/app/feed_ingest_loop.cpp has a third, separately hand-written copy of the exact same split-on-'/' logic, applied to feed_binding.ore_key (also oresmd-derived).
• IR: ores.synthetic.service/ir_curve_feed.cpp's qualifier_ is currency_code + "/" + index_display_suffix(cfg), itself built from index_family=/=tenor (migrated in the ir_curve_generation_config task) – already oresmd-derived, no splitting needed since IR publishes qualifier directly rather than a combined key string. ores.marketdata.service/app/curve_feed_ingest_loop.cpp is a pure pass-through: it copies tick->qualifier=/=tick->point_id verbatim, no generation logic of its own.
• point_id for IR ticks comes from ir_curve_resolved_entry.point_id (a tenor code, e.g. "3M"), sourced from ir_curve_template_resolver.cpp. Checked oresmd's own point field (market_data_identifier.hpp): it is a plain std::optional<std::string> with worked examples like "5y" – a bare tenor code already is exactly the oresmd point shape. No transformation needed; this was already correct before this task started.
• No hand-entry UI: ores.marketdata.market_series.org has :ores.cpp.qt.enabled: false – confirmed no Detail Dialog exists for this entity at all. market_series rows are only ever auto-created by feed_ingest_loop.cpp=/=curve_feed_ingest_loop.cpp=/=import_service.cpp, never hand-typed.
• Out of scope: ores.ore.core/src/market/market_data_parser.cpp and series_key_registry.cpp parse real, externally-authored ORE market-data files/keys (e.g. a vendored market.txt) into type/metric/qualifier/point_id – the reverse direction from this story's oresmd URI generation work, and import_service.cpp's own find_or_create_series=/observation-write logic is downstream of that parser, not a free-text-vs-oresmd generation choice. This is explicitly the separate [[id:A202712C-9CFD-48DC-9014-40BA5A9408C6][Build ORE curve-config and market-data XML round-trip via oresmd]] task's domain, per this story's own =* Decisions; not touched here.

Add a shared utility (likely ores::marketdata::core::oresmd_projections::split_projected_key or a small free function alongside it in ores.marketdata.core/oresmd/) that splits a TYPE/METRIC/QUALIFIER... string (the shape to_quote_key=/=to_curve_key etc. already produce) into {series_type, metric, qualifier}, with the same "qualifier absorbs every remaining slash-segment" behaviour the three existing ad-hoc copies share. Replace:

• ores.synthetic.service/feed_controller.hpp' two copies,
• ores.marketdata.service/app/feed_ingest_loop.cpp's copy,

with calls to the shared function. Add unit tests for the shared function itself (trivial, pure, easily testable unlike the NATS-dependent call sites).

Update this task's * Acceptance checkboxes with findings rather than code changes for FX qualifier generation, IR qualifier/point_id generation, and the no-hand-entry-UI confirmation – these were already correct going in, a direct consequence of the two prior tasks on this story having done their job properly.

Full compass build + ctest; add/extend a small unit test suite for the new shared split-key function (pure, no NATS dependency, same "extract for testability" pattern used for ir_curve_feeds_conflict=/=select_vintage_anchor_entry).