Tests: cover the security-critical paths of the crypto and platform layers.

[bernardladenthin]

What

Adds 14 test blocks to the existing self-test suites. No production code is touched and no
existing test is modified — 1133 lines, additions only, across 6 files:

File	Blocks added
src/Volume/EncryptionTest.cpp / .h	8 — API rejection paths, hash wrappers, KDF selection, volume-header rejection, password handling, keyfile mixing, VolumeInfo round trip, volume layouts
src/Platform/PlatformTest.cpp / .h	4 — Buffer/SecureBuffer/Memory, StringConverter, file operations, exception transport
src/Main/UserInterface.cpp / .h	2 — random number generator, MountOptions round trip

The tests target paths the existing known-answer tests never reach, because those only ever
take the success path: rejection of bad parameters, uninitialised state, out-of-range access,
and the serialisation round trips that cross the IPC boundary to the privileged core service.

Merge order

Please merge #1844 first. Both PRs touch src/Volume/EncryptionTest.{cpp,h}; the changes are
independent, but #1844 declares its method before TestCiphers while this branch inserts
after it. A plain git apply of #1844 onto this branch fails on the header context, while
git apply -3, a merge or a rebase all succeed. Merging #1844 first avoids the question. I
will rebase this branch afterwards.

Coverage

Measured A/B in a single binary — same build, the new test calls commented out for the
baseline — using gcov. Base is master @ b48e31f5 (1.26.29).

File	before	after
Volume/Hash.cpp	0.00 %	96.88 %
Volume/VolumeLayout.cpp	0.00 %	95.14 %
Volume/VolumePasswordCache.cpp	0.00 %	94.44 %
Core/RandomNumberGenerator.cpp	0.00 %	84.51 %
Volume/Keyfile.cpp	0.00 %	58.95 %
Volume/VolumeInfo.cpp	3.16 %	73.68 %
Platform/FileCommon.cpp	23.26 %	86.05 %
Core/MountOptions.cpp	27.12 %	85.59 %
Platform/Memory.cpp	41.18 %	91.18 %
Volume/VolumePassword.cpp	45.83 %	100.00 %
Volume/Pkcs5Kdf.cpp	50.48 %	70.48 %
Volume/Cipher.cpp	73.24 %	83.57 %
Platform/Buffer.cpp	75.76 %	83.84 %
Volume/EncryptionAlgorithm.cpp	78.28 %	93.43 %
Volume/VolumeHeader.cpp	81.37 %	83.01 %
Volume/EncryptionModeXTS.cpp	91.39 %	93.38 %
Platform/StringConverter.cpp	43.29 %	61.90 %

17 files improved, five of them starting from zero.

Coverage was produced without modifying any build file, using the existing hooks in
src/Makefile:

make NOGUI=1 NOTEST=1 \
     TC_EXTRA_CFLAGS="--coverage -O0" \
     TC_EXTRA_CXXFLAGS="--coverage -O0" \
     TC_EXTRA_LFLAGS="--coverage"

A note for anyone repeating this: use gcov directly. Both lcov and gcovr mis-attribute
results here, because each subproject is built through make -C and the working directory
recorded in the .gcno files cannot be resolved afterwards. lcov reported one file at 8.7 %
where the true figure is 91.4 %, and produced merged totals lower than the individual runs.

Detection proof

Coverage alone does not show that a test would catch anything, so every block was verified by
injecting a defect and confirming that the new test itself reports it. 27 injected defects in
total, among them:

Injected defect	Reported by
Buffer::Erase no longer wipes	BufferTest
Buffer::GetRange skips its bounds check	BufferTest
Memory::Compare inverts its ordering	BufferTest
keyfile contributes nothing to the password pool	TestKeyfileApplication
base password dropped from the keyfile pool	TestKeyfileApplication
empty keyfile silently accepted	TestKeyfileApplication
MountOptions loses the protection mode in transit	TestMountOptionsSerialization
password truncated in transit	TestPasswordHandling
password cache grows without bound	TestPasswordHandling
password cache stops deduplicating	TestPasswordHandling
RNG returns constant data	TestRandomNumberGenerator
RNG usable before Start()	TestRandomNumberGenerator
oversized RNG request not rejected	TestRandomNumberGenerator
File::Copy truncates the copy	FileTest
ReadCompleteBuffer accepts short reads	FileTest
exception subject dropped on the IPC path	ExceptionTransportTest
cipher accepts block operations before a key is set	TestEdgeCases
SetKey accepts a wrong-length key	TestEdgeCases
hash accepts empty input	TestHashClasses
KDF lookup matches any name	TestKdfSelection
volume layout reports a data area past the end of the host	TestVolumeLayouts

One side effect worth mentioning: the random number generator carries its own pool-mixing
known-answer self-test inside Start(). That self-test previously ran only when the graphical
or the text interface started the generator — --test never started it, so it never ran under
CI. It does now.

Verification environment

Linux — Docker ubuntu:24.04, GCC 13, distribution wxWidgets 3.2, make NOGUI=1. The
repository is mounted read-only and copied into the container before building, so no build
artefacts touch the working tree. The self-tests were run twice to confirm repeatability, and
the temporary files created by the keyfile and file tests are removed even when an assertion
fires.

Windows — Visual Studio 2022 17.14.37, MSVC 14.44.35228, Windows SDK 10.0.26100.0,
WDK 10.0.26100. Each affected translation unit was compiled in user mode at /W4 and in
kernel mode at /W4 /WX, matching the TreatWarningAsError setting in Driver.vcxproj.

None of the six changed files appears in any .vcxproj, so the Windows build cannot be
affected by this change. The compile checks were run regardless.

Limitations

• Volume/VolumeHeader.cpp moves by only 1.6 pp. Its remaining uncovered part is the
  decrypt-attempt loop, which needs a real encrypted volume — out of scope for a unit test.
• Volume/Keyfile.cpp stops at 59 %. The remaining branches are the security-token and
  directory-enumeration paths, which need a token or a populated directory.
• The repository has no Windows CI job, so the Windows half of these tests runs only when
  Format, Mount or ExpandVolume actually start, not automatically.

AI assistance

These tests were written with the assistance of Anthropic's Claude Opus. Every claim in this
description — the coverage figures, the injected-defect results, the build and warning output
— was produced by running the builds and measurements described above rather than estimated.
The changes are additive and confined to test code; review of the assertions themselves is of
course still welcome.

[bernardladenthin]

Please merge #1844 before this PR. Both touch src/Volume/EncryptionTest.{cpp,h}, and while
the changes are independent (different functions, no overlapping assertions), applying them in
the other order requires a three-way merge — a plain git apply of #1844 will fail against
this branch's header. Merging #1844 first avoids that entirely; I will rebase this branch on
the updated master afterwards.