Allow # in file names

[davidhassell]

Fixes #356

Essentially targets uses of uritools.urisplit and uritools.uricompose to account for URI fragments. Makes use of cfdm.abspath and cfdm.dirname where possible, as these now deal with it appropriately.

[sadielbartholomew]

Sensible fix that works well from local testing (including downstream in cf-python), but I would like a dedicated test added to cover it - should be as simple as reading in an existing field (list), writing it out with a name with 2 or 3 hash characters in and then checking it re-reads - unless there's some reason that complicates like temp files not liking the '#' in the name?

(I'm guessing having a '#' in the name of a dataset does happen in practice - else I'd be tempted to argue the users should rename such files to have standard characters only rather than us fixing a weird edge case that fails for good reason... 😃 )

[davidhassell]

Test added: c5cb747

[sadielbartholomew]

Thanks for adding the test. It seems sufficiently robust, so all good now - please merge.

[davidhassell]

(I'm guessing having a '#' in the name of a dataset does happen in practice - else I'd be tempted to argue the users should rename such files to have standard characters only rather than us fixing a weird edge case that fails for good reason... 😃 )

Unfortunately so. Charlotte's use case was from FAMOUS model output.