Task: Implement Hotfix: the nightly Valgrind run reports uninitialised members and library leaks

Take the four suites the nightly dynamic analysis reports defects for to zero: ores.trading.api.tests (4), ores.refdata.core.tests (3329), ores.ore.core.tests (99) and ores.analytics.core.tests (214). Fix the defects that are ours and suppress the allocations that are the libraries'.

• Every uninitialised-value defect the four suites reported is gone, and the member that caused it is initialised by its model's :default_value:.
• The 3329 still-reachable leak records for ores.refdata.core.tests are suppressed by rules anchored on the allocating library, each with the reason it is not a leak.
• A nullable column whose :cpp_type: is a bare value type can no longer be merged without a declared default: org_loader refuses it.
• The full test suite passes, the codegen suite passes, and the sweep over every catalogue component reports no file this branch changed as drifted.

1. Reproduce each reported defect locally with --track-origins=yes so the member and the site that leaves it unset are named, rather than inferred from a truncated CDash page.
2. Fix each at the model: :default_value: on the column, then regenerate the component.
3. Census the rest of the class so the size of what is left is on record.
4. Extend build/valgrind/custom.supp for the library allocations, and add the codegen gate for the nullable shape.
5. Re-run the four binaries under valgrind until the logs are empty, and run the whole suite and the drift gates before raising the pull request.

Every number this task claims about a tree, with the command that produced it and the commit it was measured at.

• The CDash pages saved under tmp/ are truncated at about a megabyte, so the analytics page listed 64 of the 214 defects and the first fix left 146 behind in a second entity. The complete set came from re-running each binary locally, which is also what proves the fix.
• The uninitialised values are all created by a stack frame in the test or the generated generator, and read in a write path: sqlgen stringifies every entity member to build the INSERT, so an entity built by a generated generator that skips a column is read whole.
• bond_trs::funding_rate and bond_repo::repo_rate are nullable: true with a bare double, so the generated entity wraps them in std::optional and the generated mapper maps zero to nullopt. The declared default of 0.0 is that convention, not a value the domain intends.
• The full suite passes on a recreated database, and only on a recreated database: ores.iam.core.tests is not idempotent. Its first run leaves three seed profiles open, and the second run fails three cases on seed_profiles_code_uniq_idx. That is a pre-existing defect of the tests rather than of this change, and it is filed as The iam seed-profile eventing tests only pass once per database. It also cost this task two verification runs: the residue was misread at first as damage the valgrind runs had done.
• The gate refuses the nullable shape rather than synthesising a default, because the value that means "absent" is the model's to state. For the non-nullable shape the same choice is open and is filed as a capture; this task declares the defaults the reported instances needed and records the census, so the decision is taken with the size known.

The four suites the nightly reported are clean under valgrind: each log holds no error and no leak record. Twenty-four model columns declare the default their generated member needs, the nullable shape that caused several of them is refused by org_loader, and the still-reachable library allocations are suppressed with the reason each is not a leak.

The full suite passes 89 of 89 on a recreated database, the codegen suite passes 1001 tests, and the sweep over the catalogue reports no file this branch changed as drifted. What is left of the class is filed as a capture with the census: 87 non-nullable columns still generate an indeterminate member.