Use local instances of MagneticFieldGridReader in MagneticFieldSvc

Fixes a minor annoyance with the MagneticFieldSvc which is messages sent from it were not respecting changes made to the message formatting. This was due to the timing of when the Reader instances where constructed. This MR fixes this by instead of having a single cache instance, use local instances instead. As the reader has no cached state, and is cheap to construct, this has no overhead.

• [2018-05-11 11:36] Validation started with lhcb-gaudi-merge#513
• [2018-05-12 00:11] Validation started with lhcb-lcg-dev3#522
• [2018-05-12 00:12] Validation started with lhcb-clang-test#949
• [2018-05-12 00:13] Validation started with lhcb-lcg-dev4#522
• [2018-05-12 00:14] Validation started with lhcb-gaudi-head-py3#151
• [2018-05-12 00:17] Validation started with lhcb-gaudi-head#1864
• [2018-05-12 00:30] Validation started with lhcb-tdr-test#148
• [2018-05-12 00:47] Validation started with lhcb-head#1850
• [2018-05-12 08:20] Validation started with lhcb-gaudi-merge#514
• [2018-05-12 11:37] Validation started with lhcb-gaudi-merge#515

changed the description

@jonrob is this relevant also for run2-patches?

merged

mentioned in commit 9edbaaf5

Yes, I think it should cherry pick fine to run2-patches, and if it does applying it there as well makes sense.

mentioned in commit 450d51fd

mentioned in merge request !1319 (merged)