[GTiff] GTiffOddBitsBand::IWriteBlock crashes with anything different from eDataType=GDTByte

In trunk:

- in r14840, I've commited a fix to enable updating data with NBITS >= 9 and NBITS <= 32, which crashed before. The float case with nBitsPerSample < 32 is not handled, but a proper error is returned. In the process of fixing that, I've also decided to remove the implementations of GTiffBitmapBand::IReadBlock/IWriteBlock to use GTiffOddBitsBand instead (dealt as special cases to keep performance). That made it easier to fix a subtle bug in round-tripping IReadBlock/IWriteBlock which can be shown with the following script :

import gdal
import struct

oldSize = gdal.GetCacheMax()
gdal.SetCacheMax(0)

# NBITS=2 will also fail with GDAL 1.5
ds = gdal.GetDriverByName("GTiff").Create("temp.tif", 10, 10, 1, options = [ 'NBITS=1' ])
ds.GetRasterBand(1).Fill(0)

ds2 = gdal.GetDriverByName("GTiff").Create("temp2.tif", 10, 10, 1)
ds2.GetRasterBand(1).Fill(1)
ones = ds2.ReadRaster(0, 0, 10, 1)

data = ds.ReadRaster(0, 0, 10, 1)

ds.WriteRaster(0, 0, 10, 1, ones)

# This will exhaust the cache
ds3 = gdal.GetDriverByName("GTiff").Create("temp3.tif", 10, 10, 1)
ds3.GetRasterBand(1).Fill(1)

# We expect (1, 1, 1, 1, 1, 1, 1, 1, 1, 1)
data = ds.ReadRaster(0, 0, 10, 1)
got = struct.unpack('B' * 10, data)
for i in range(len(got)):
    if got[i] != 1:
        print 'fail'

gdal.SetCacheMax(oldSize)

It is now possible to create and translate files with GDT_UInt16 where NBITS>=9 and NBITS<=15, and with GDT_UInt32 where NBITS>=17 and NBITS<=32

All that is tested in r14841

- in branches/1.5 in r14842, I've just fixed the tile of this bug, that is to say to error properly when NBITS > 8.