From the DESeq2 vignette:

txi <- tximport(files, type="salmon", tx2gene=tx2gene)

ddsTxi <- DESeqDataSetFromTximport(countData = txi, colData = samples, design = ~ condition)

Q1. Is it possible to normalise a gene expression matrix (produced by Salmon and tximport) using DESeq2 method without sample information?

The goal here is to obtain a DESeq normalised expression matrix.

Q2. As an alternative, would it be coherent to normalise a matrix of raw read counts manually by the method of median of ratios (described here)?

Q3. Is it correct to say that RLE is the DESeq2 normalisation method, whereas the median of ratios was the previous DESeq method?

A1. Yes. Try giving a trivial dataframe for colData and a trivial design (like ~1) for design.

A2. I'd rather use the estimateSizeFactorsForMatrix for that alternative.

A3. No.
The normalization methods of DESeq2 and DESeq are (up to implementation details) the same.