The syntax suite (scripts/test_syntax.sh) is a merge-gating test on all
platforms but contributed no coverage: it drives the res_parser binary,
which is instrumented (it carries the bisect_ppx stanza and links the
instrumented compiler/syntax lib) and is invoked from
_build/install/default/bin, the same pattern as the analysis tests.

Run it under BISECT_FILE so compiler/syntax is no longer a blind spot.
ROUNDTRIP_TEST=1 matches Linux CI and also exercises the printer paths.

The reanalyze tests invoke rescript-tools through `dune exec`, so they
ran an uninstrumented binary and produced no coverage (and a plain
`dune exec` would rebuild the shared _build artifacts uninstrumented).

Set DUNE_INSTRUMENT_WITH=bisect_ppx (the env-var form of
`--instrument-with`) for the analysis suite so that `dune exec` matches
the instrumentation config coverage-build already used. dune finds the
build up to date, runs the instrumented binary, and reanalyze now
contributes coverage for analysis/reanalyze without any test-script
changes. Switch the analysis line from test-analysis-binary to the full
test target to pull reanalyze in.

The DUNE_INSTRUMENT_WITH change alone routed the reanalyze tests to an
instrumented rescript-tools, but the reanalyze library itself declared
no instrumentation backend, so its own 58 source files recorded zero
coverage (only the analysis/tools code it links was counted).

Add the (instrumentation (backend bisect_ppx)) stanza — matching every
other analysis/* and tools/* library — so analysis/reanalyze/src is
measured. Verified locally: reanalyze goes from 0 to ~61% line
coverage in the report.

The stanza is a no-op for normal builds (it only activates under
--instrument-with / DUNE_INSTRUMENT_WITH).