Generates a test case for every known lens from canonCsLensType, that first
sets the corresponding lens metadata and then verifies that exiv2 maps it
to the expected lens description. Only metadata fields that are relevant
for lens identification are modified.

There is no need to handle tests on Windows and Unix differently here.
Always using a shell allows for more flexibility when writing tests.

(rebased by hassec)

Lenses that have the exact same ID, focal length and aperture as some other
lens that comes earlier in the list (and thus always wins):
* 137, "Tamron SP 17-50mm f/2.8 XR Di II VC"
* 137, "Tamron SP 24-70mm f/2.8 Di VC USD"
* 161, "Sigma 28-70mm f/2.8 EX"
* 173, "Sigma 180mm EX HSM Macro f/3.5"
* 180, "Zeiss Milvus 50mm f/1.4"
* 183, "Sigma 150-600mm f/5-6.3 DG OS HSM | S"
* 254, "Tamron SP 90mm f/2.8 Di VC USD Macro 1:1 F004"
* 254, "Tamron SP 90mm f/2.8 Di VC USD Macro 1:1 F017"

Lenses that share their IDs with other lenses, but have no or an
unsupported focal length:
* 33, "Voigtlander or Carl Zeiss Lens"
* 131, "Sigma 4.5mm f/2.8 EX DC HSM Circular Fisheye"

Fixes some small inconsistencies, so that all lenses use the same format,
that is also shared with other lens databases such as lensfun:
* Always prefix aperture with f/
* Never add .0 to aperture
* Always add mm to focal length
* Always use | A for Sigma Art lenses

The mathematical calculation of fnumbers does not always match the expected
values: For example for f/3.5 the precise mathematical value is 3.564...,
which gets rounded to 3.6. Fix this special case by returning a value
closer to the expected value.

When searching for the Tamron lens, only the string "300mm" is searched in
the lens description, which also happens to be present for the Canon lens.
Since the Canon lens comes first in the list, it wins. Fix this issue by
prefixing the search string with a single space so it always has to match
the full focal length specification.

Lenses with and without a TC may share the same lens ID. Prefer entries
that explicitly mention the TC.

This commit does some restructuring to make common utils available
for future similar test for other brands

For each lens, its test target is now defined as the list of all lenses
which are possible given that lenses exif values.

…biguous lenses

If multiple choices are possible they are now all reported. This
behaviour is now the same as it is in exiftool.

All lenses are tested in the new test_canon_lenses.py test