[magnusp]

Summary

LibsqlResultSet.getBytes(int) encoded REAL values with Double.toString, so 0.0001 came back as 1.0E-4 and disagreed with getString. REAL values now go through the existing sqliteRealText helper, so getBytes returns the same SQLite canonical text bytes as getString (0.0001, 10000000.0, 1.0e+20).

Contract. Hrana 3 is the source of truth for transport, values, and metadata. java.sql is the source of truth for the JDBC surface. xerial/sqlite-jdbc 3.53.4.0 is a behavioral reference only. Hrana Value has no text form for a float, so the driver formats locally, targeting the text the server produces for CAST(value AS TEXT). That is a SQLite conversion rule, not a protocol field and not an xerial quirk.

Fixes #74

Changes

• LibsqlResultSet.getBytes(int) routes Value.Float through sqliteRealText. One-line change; the BLOB, TEXT, INTEGER, and NULL branches are untouched.
• LibsqlResultSetRealTextTest pins getBytes against the canonical text for all 11 existing cases, including the three issue values, and asserts equality with getString bytes.

Evidence

Red at trunk (277ef8d), ./mvnw -pl libsql-jdbc test -Dtest=LibsqlResultSetRealTextTest: 10 of the 11 new parameters fail. Those include 0.0001 -> 1.0E-4, 10000000.0 -> 1.0E7, and 1e20 -> 1.0E20. 0.0 passes because Double.toString(0.0) already produces 0.0.

Green at head (c0fae0d):

Command	Result
./mvnw -pl libsql-jdbc test -Dtest=LibsqlResultSetRealTextTest	25 tests, 0 failures
./mvnw -pl libsql-jdbc test -Dtest='LibsqlResultSetRealTextTest,LibsqlResultSetTest'	43 tests, 0 failures
./mvnw -pl libsql-jdbc test	189 tests, BUILD SUCCESS
./mvnw spotless:check	BUILD SUCCESS
./mvnw test	BUILD SUCCESS
./mvnw -pl libsql-jdbc -am package -DskipTests	driver jar built
./mvnw -pl libsql-conformance -am verify -Dit.test=org.sqlite.ResultSetTest	21 tests, BUILD SUCCESS
./mvnw -pl libsql-conformance -am verify	115 tests, 1 skipped (pre-existing InsertQueryTest skip), no new divergence
./mvnw -pl libsql-conformance -am -Pxerial-control verify	100 tests, 1 skipped, BUILD SUCCESS

BLOB and TEXT getBytes are unaffected: the diff touches only the Value.Float branch, and the vendored ResultSetTest, which exercises getBytes on INTEGER and TEXT, stays green.

Perf probe, wall clock of ./mvnw -pl libsql-jdbc test, 3 interleaved runs per side after one warmup: trunk median 2.17 s, head median 2.20 s, head range 2.12 to 3.44 s. No measurable difference, and the metric is dominated by Maven startup.

Divergence

None. This removes an internal divergence between getBytes and getString for REAL. The xerial control lane completed with no failures, and the vendored ResultSetTest stays green.

Base 277ef8d / head c0fae0d.